Choose AI / Free and local

The best free AI on your own computer

Open-source models: you download them and use them without paying, without limits and without sending your data to anyone.

Our picks

The option we recommend today.

Our pick

Qwen3.8 27B

Alibaba

Score 34/100

Because it's the smartest AI that runs on a computer with a single graphics card (it needs about 20.0 GB of video memory), with no payment and no limits.

You'll find it inDownload it (open source) · How to use it

Free (open source)

Open sourceProvenWidely used

The best open source, by task

Each one with the video memory (VRAM) it needs.

Your computer By default, a 24 GB graphics card. Change it and the lists update.

Your graphics card memory (VRAM)

Other

Computer RAM GBFree disk space GB Mac with Apple Silicon (unified memory)

Don't know how much VRAM you have? On Windows: Task Manager → Performance → GPU (“Dedicated GPU memory”). On a Mac: Apple menu → About This Mac (“Memory”).

How we work it out: if you have a graphics card, the model has to fit in its memory (VRAM). On a Mac with Apple Silicon we count 70% of the unified memory. Without a graphics card, or if it doesn't fit in it, it can run from RAM (we count 70%), but it will be slow. The memory each model needs is its official requirement if there is one; otherwise, the file of its lightest common version (usually reduced to 8 or 4 bits) + 10% + 2 GB. If we don't know how much memory it needs, we don't count it.

These details are saved only in this browser and are never sent anywhere.

General

Qwen3.8 27Bneeds ~20 GB, you have 24 GB of VRAM · Q4 version · ~16.5 GB file · How to use it
Score 34/100
Muse Glimmerneeds ~19 GB, you have 24 GB of VRAM · Q4 version · ~15.9 GB file · How to use it
Score 18/100
Gemma 4 31Bneeds ~22 GB, you have 24 GB of VRAM · Q4 version · ~18.3 GB file · How to use it
Score 15/100

Coding

Qwen3.8 27Bneeds ~20 GB, you have 24 GB of VRAM · Q4 version · ~16.5 GB file · How to use it
Score 6/100

Images

Create images

Qwen-Image-2.1needs ~10 GB, you have 24 GB of VRAM · FP8 version · ~7.1 GB file · How to use it
Score 56/100
Ideogram 4.0 (Quality)needs ~12 GB, you have 24 GB of VRAM · FP8 version · ~9.3 GB file · How to use it
Score 51/100
Ideogram 4.0 Fast (Quality)needs ~22 GB, you have 24 GB of VRAM · BF16 version · ~18.6 GB file · How to use it
Score 46/100

Video

Create videos

LTX-2.3 Proneeds ~16 GB, you have 24 GB of VRAM · official requirement · ~14.3 GB file · How to use it
Score 36/100
LTX-2.3 Fastneeds ~16 GB, you have 24 GB of VRAM · official requirement · ~14.3 GB file · How to use it
Score 30/100

General

MiMo-V2.6-Proneeds 625 GB of VRAM · How to use it
Score 46/100
GLM-5.3needs 833 GB of VRAM · How to use it
Score 45/100
Kimi K3needs 1719 GB of VRAM · How to use it
Score 44/100
GLM-5.3-Flashneeds 363 GB of VRAM · How to use it
Score 42/100
Qwen3.8 2.4T A95Bneeds 5384 GB of VRAM · How to use it
Score 40/100

Coding

GLM-5.3needs 833 GB of VRAM · How to use it
Score 42/100
MiMo-V2.6-Proneeds 625 GB of VRAM · How to use it
Score 35/100
GLM-5.3-Flashneeds 363 GB of VRAM · How to use it
Score 33/100
DeepSeek V4.1 Flashneeds 563 GB of VRAM · How to use it
Score 27/100
Qwen3.8-Flash-Nextneeds 124 GB of VRAM · How to use it
Score 25/100

Images

Create images

Qwen-Image-2.1needs 10.0 GB of VRAM · How to use it
Score 56/100
Ideogram 4.0 (Quality)needs 12.0 GB of VRAM · How to use it
Score 51/100
Ideogram 4.0needs 12.0 GB of VRAM · How to use it
Score 49/100
FLUX.2 [dev]needs 73.0 GB of VRAM · How to use it
Score 48/100
Qwen Image Max 2512needs 47.0 GB of VRAM · How to use it
Score 48/100

Video

Create videos

MiniMax H3needs 25.0 GB of VRAM · How to use it
Score 95/100
LTX-2.5 Fastneeds 32.0 GB of VRAM · How to use it
Score 46/100
LTX-2.5 Proneeds 32.0 GB of VRAM · How to use it
Score 45/100
LTX-2.3 Proneeds 16.0 GB of VRAM · How to use it
Score 36/100
LTX-2.3 Fastneeds 16.0 GB of VRAM · How to use it
Score 30/100

Data zone

For the data nerds

Open-weight models by category: quality vs. price chart and full table with parameters, memory (VRAM), variants and licences. The “✓ Fits on your computer” mark = fits on a 24 GB graphics card (the rule above with the default values).

Sources publish API prices in dollars. With the €/$ button at the top you choose the currency; in euros they are converted at €1 = $1.1206 (ECB reference rate of October 9, 2026).

17 open-source models and variants in general · 5 fit on your computer

Intelligence vs cost per task (open only)

Up = better · left = cheaper · the dashed line is the efficient frontier.

14 models · 4 on the frontier · hover over or tap a point to see its name and values

ClosedOpen weightsEfficient frontier

Swipe to see the whole chart

Price with the cheapest API that offers it; on your computer the cost is your hardware and electricity.

Table: General (open only)

17 models and variants · click ▸ on a row to see sources, memory and notes

Sort

Model	IndexAA ⇅	Costper task ⇅	WeightsWeights⇅	MaturityMad.⇅	€/1M$/1Min / out	Paramstotal / act.	Memory ⇅VRAM · disco	License
MiMo-V2.6-Pro Xiaomi	46	€0.12$0.13	OpenOpen	NewNew ⬇ 93k/mo HF · 4 rankingsModerately usedModerately used	€0.39$0.43 / €0.78$0.87	1.0T / 42B	625 GB approx. ▣ 1 node 8 GPU disk 566 GB 1 node 8× Blackwell (2 nodes on Hopper)	MIT
Featured: ★ Best open (and cheap) · on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Released 18 days ago · 93k HF downloads/mo · on OpenRouter · 4 leaderboards · Adoption: Moderately used (⬇ 93k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-21; data from 2026-10-09) Score: Best open by index (Sep 21). Memory: disk 566 GB (FP4/FP8 mixed) · Min. VRAM 625 GB approx. · fits on: 1 node 8× Blackwell (2 nodes on Hopper) Official FP4/FP8 566 GB Requirement: Model card: vLLM with tensor-parallel 8 (SGLang tp 16). GenAI Brief: ~1 TB in FP8; one node of 8 Blackwell GPUs. source License: MIT Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
GLM-5.3 (max) Z AI	45	€1.79$2.01	OpenOpen	ProvenProven OpenRouter top 20 · #4 · 5 rankingsWidely usedWidely used	€1.25$1.40 / €3.93$4.40	753B / 40B	833 GB approx. ▣ 1 node 8 GPU disk 756 GB 1 node 8× H200 (FP8)	GLM-5.3 (condition >$10,000M MaaS)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.5 M HF downloads/mo · top 20 OpenRouter · 5 leaderboards · 52 days · Adoption: Widely used (OpenRouter top 20 · #4 · 5 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, AA Coding Agents, LMArena Text, LMArena WebDev; released: 2026-08-18; data from 2026-10-09) Memory: disk 756 GB (FP8) · Min. VRAM 833 GB approx. · fits on: 1 node 8× H200 (FP8) Official FP8 756 GB Requirement: GenAI Brief: ~750 GB in FP8; fits on an 8× H200 node. source Repo zai-org/GLM-5.3-FP8 returns 401 (not accessible). License: GLM-5.3 (condition >$10,000M MaaS) Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Kimi K3 (max) Moonshot AI	44	€1.78$2.00	OpenOpen	ProvenProven OpenRouter top 20 · #6 · 5 rankingsWidely usedWidely used	€2.68$3.00 / €13.39$15.00	2.8T / 104B	1719 GB approx. ▣ Multi-node disk 1561 GB Multi-node on Hopper; high-memory Blackwell node	Kimi K3 (attribution and MaaS agreement >$20M)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.2 M HF downloads/mo · top 20 OpenRouter · 5 leaderboards · 85 days · Adoption: Widely used (OpenRouter top 20 · #6 · 5 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, AA Coding Agents, LMArena Text, LMArena WebDev; released: 2026-07-16; data from 2026-10-09) Memory: disk 1561 GB (MXFP4/U8) · Min. VRAM 1719 GB approx. · fits on: Multi-node on Hopper; high-memory Blackwell node Official MXFP4 1561 GB GGUF UD-IQ1_S (community) 594 GB Requirement: GenAI Brief: ~1.4 TB in MXFP4; multi-node on Hopper. source License: Kimi K3 (attribution and MaaS agreement >$20M) Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
GLM-5.3-Flash Z AI	42	€0.23$0.25	OpenOpen	ProvenProven OpenRouter top 20 · #11 · 4 rankingsWidely usedWidely used	€0.13$0.15 / €0.45$0.50	320B / 18B	363 GB approx. ▣ 1 node 8 GPU disk 328 GB 1 node 8× H100 (FP8)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 6.4 M HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 44 days · Adoption: Widely used (OpenRouter top 20 · #11 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-26; data from 2026-10-09) Memory: disk 328 GB (FP8) · Min. VRAM 363 GB approx. · fits on: 1 node 8× H100 (FP8) Official FP8 328 GB GGUF Q8_0 341 GB GGUF UD-IQ1_S 93.1 GB Requirement: GenAI Brief: Flash fits comfortably on 8× H100. source License: MIT Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Qwen3.8 2.4T A95B Alibaba	40	€1.92$2.16	OpenOpen	ProvenProven ⬇ 33k/mo HF · 2 rankingsWidely usedWidely used	€1.78$2.00 / €5.35$6.00	2.4T / 95B	5384 GB approx. ▣ Multi-node disk 4892 GB Multi-node (does not fit on a Hopper node)	Qwen3.8-Max (commercial terms)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 33k HF downloads/mo · on OpenRouter · 2 leaderboards (AA Intelligence Index, AA Terminal-Bench 4.0) · 58 days · Adoption: Widely used (⬇ 33k/mo HF · 2 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0; released: 2026-08-12; data from 2026-10-09) Memory: disk 4892 GB (BF16) · Min. VRAM 5384 GB approx. · fits on: Multi-node (does not fit on a Hopper node) Official BF16 4892 GB Requirement: GenAI Brief: does not fit on a Hopper node at any published precision. source License: Qwen3.8-Max (commercial terms) Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Qwen3.8-Flash-Next Alibaba	40	€0.33$0.37	OpenOpen	ProvenProven ⬇ 1.6M/mo HF · 3 rankingsWidely usedWidely used	€0.13$0.15 / €0.42$0.47	180B / 6B	124 GB approx. ▣ 2–4 GPU disk 360 GB 1× H200 (141 GB) or 2× 80 GB (approx.)	Qwen Community 1.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.6 M HF downloads/mo · 3 leaderboards · 44 days · Adoption: Widely used (⬇ 1.6M/mo HF · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena WebDev; released: 2026-08-26; data from 2026-10-09) Memory: disk 360 GB (BF16) · Min. VRAM 124 GB approx. · fits on: 1× H200 (141 GB) or 2× 80 GB (approx.) Official BF16 360 GB GGUF UD-Q4_K_XL 111 GB GGUF UD-IQ1_S 72.5 GB 180B total / 6B active: in Q4 it runs on 2× 80 GB or a Mac with 128 GB+ (approx.). License: Qwen Community 1.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
DeepSeek V4.1 Flash (max) DeepSeek	40	€0.24$0.27	OpenOpen	ProvenProven OpenRouter top 20 · #3 · 4 rankingsWidely usedWidely used	€0.27$0.30 / €1.07$1.20	552B / 16B	563 GB approx. ▣ 1 node 8 GPU disk 510 GB 1 node 8 GPU (FP8)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.3 M HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 29 days · Adoption: Widely used (OpenRouter top 20 · #3 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-10; data from 2026-10-09) Score: The fastest of the big ones (~216 tok/s). Memory: disk 510 GB (FP8/INT8) · Min. VRAM 563 GB approx. · fits on: 1 node 8 GPU (FP8) Official FP8 510 GB GGUF Q2 (antirez) 366 GB Requirement: GenAI Brief: ~550 GB in FP8; one 8-GPU node. source License: MIT Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
MiMo-V2.6-Flash Xiaomi	38	€0.056$0.062	OpenOpen	ProvenProven OpenRouter top 20 · #8 · 4 rankingsWidely usedWidely used	€0.12$0.14 / €0.25$0.28	309B / 15B	186 GB approx. ▣ 2–4 GPU disk 173 GB 4× GPU 80 GB (TP 4)	MIT
Featured: ★ Ultra-cheap open · on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 56k HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 18 days · Adoption: Widely used (OpenRouter top 20 · #8 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-21; data from 2026-10-09) Memory: disk 173 GB (FP4/FP8 mixed) · Min. VRAM 186 GB approx. · fits on: 4× GPU 80 GB (TP 4) Official FP4/FP8 173 GB GGUF MXFP4 167 GB Requirement: Model card: vLLM example with tensor-parallel 4. source License: MIT Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Qwen3.8 27B (xhigh) Alibaba	34	€0.90$1.01	OpenOpen	ProvenProven ⬇ 6.8M/mo HF · 4 rankingsWidely usedWidely used	€0.45$0.50 / €2.68$3.00	27B / 27B (dense)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 55.6 GB 1× 80 GB (BF16) · 1× 24–32 GB (Q4)	Apache 2.0
Featured: ★ Fits on 1 GPU Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 6.8 M HF downloads/mo · on OpenRouter · 4 leaderboards · 56 days · Adoption: Widely used (⬇ 6.8M/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-14; data from 2026-10-09) Score: The best one that fits on a single GPU. Memory: disk 55.6 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 80 GB (BF16) · 1× 24–32 GB (Q4) Official BF16 55.6 GB Official FP8 30.9 GB GGUF UD-Q4_K_M 16.5 GB Requirement: GenAI Brief: 55.6 GB BF16 fits on an 80 GB GPU without quantizing; the default «single-GPU» model. source License: Apache 2.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
MiniMax-M3 MiniMax	29	€0.45$0.51	OpenOpen	ProvenProven ⬇ 204k/mo HF · 4 rankingsWidely usedWidely used	€0.27$0.30 / €1.07$1.20	428B / 23B	284 GB approx. ▣ 1 node 8 GPU disk 854 GB 1 node 8× H100 (8 bits)	MiniMax Community (commercial use with notice)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 204k HF downloads/mo · on OpenRouter · 4 leaderboards · from 2026-06 · Adoption: Widely used (⬇ 204k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-06-01; data from 2026-10-09) Memory: disk 854 GB (BF16) · Min. VRAM 284 GB approx. · fits on: 1 node 8× H100 (8 bits) Official BF16 854 GB Official MXFP8 444 GB GGUF MXFP4_MOE 256 GB Requirement: GenAI Brief: at 8 bits the weights fit on an 8× H100 node. source License: MiniMax Community (commercial use with notice) Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Nemotron 3 Ultra NVIDIA	23	€0.53$0.60	OpenOpen	ProvenProven OpenRouter top 20 · #14 · 3 rankingsWidely usedWidely used	€0.54$0.60 / €2.23$2.50	550B / 55B	390 GB approx. ▣ 1 node 8 GPU disk 1121 GB 4× B200/GB200 (NVFP4) or 8× H100	OpenMDW-1.1
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 760k HF downloads/mo · top 20 OpenRouter · 3 leaderboards · from 2026-06 · Adoption: Widely used (OpenRouter top 20 · #14 · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text; released: 2026-06-04; data from 2026-10-09) Score: It also publishes the training data. Memory: disk 1121 GB (BF16) · Min. VRAM 390 GB approx. · fits on: 4× B200/GB200 (NVFP4) or 8× H100 Official BF16 1121 GB Official NVFP4 352 GB Requirement: NVFP4 model card: minimum GPU 4× GB200, 4× B200, 4× GB300, 4× B300 or 8× H100. source License: OpenMDW-1.1 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Muse Glimmer (high) Meta	18	€0.051$0.057	OpenOpen	ProvenProven ⬇ 269k/mo HF · 4 rankingsWidely usedWidely used	€0.29$0.32 / €1.20$1.35	30B / n/d	19.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 59.6 GB 1× 24–32 GB (quantized)	Apache 2.0
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 269k HF downloads/mo · on OpenRouter · 4 leaderboards · 60 days · Adoption: Widely used (⬇ 269k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-10; data from 2026-10-09) Score: Open successor to Llama. Memory: disk 59.6 GB (BF16) · Min. VRAM 19.0 GB approx. · fits on: 1× 24–32 GB (quantized) Official BF16 59.6 GB GGUF Q8_0 29.6 GB GGUF UD-Q4_K_XL 15.9 GB Requirement: Meta (via GenAI Brief): runs on 24 or 32 GB consumer GPUs. source License: Apache 2.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Gemma 4 31B Google	15	self-host	OpenOpen	ProvenProven ⬇ 9.6M/mo HF · 4 rankingsWidely usedWidely used	n/d	30.7B / 30.7B (dense)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 62.5 GB 1× 24 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 9.6 M HF downloads/mo · on OpenRouter · 4 leaderboards · from 2026-04 · Adoption: Widely used (⬇ 9.6M/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-04-02; data from 2026-10-09) Score: AA gives no cost per task (price ≈ €0$0). Memory: disk 62.5 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Official BF16 62.5 GB GGUF Q8_0 33.2 GB GGUF Q4_K_M 18.3 GB Requirement: Google: 12B/26B A4B/31B designed for consumer GPUs and workstations. source License: Apache 2.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
gpt-oss-120b (high) OpenAI	12	€0.096$0.11	OpenOpen	ProvenProven ⬇ 4.0M/mo HF · 3 rankingsWidely usedWidely used	€0.13$0.15 / €0.53$0.59	117B / 5.1B	74.0 GB approx. ▣ 1× 80 GB disk 65.2 GB 1× 80 GB (H100 / MI300X)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 4.0 M HF downloads/mo · on OpenRouter · 3 leaderboards · from 2025-08 · Adoption: Widely used (⬇ 4.0M/mo HF · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text; released: 2025-08-05; data from 2026-10-09) Score: Fits on an 80 GB GPU. Memory: disk 65.2 GB (MXFP4) · Min. VRAM 74.0 GB approx. · fits on: 1× 80 GB (H100 / MI300X) Official MXFP4 65.2 GB Requirement: Model card: fits on a single 80 GB GPU (H100 or MI300X) thanks to MXFP4. source License: Apache 2.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Mistral Large 3 Mistral	9	€0.028$0.031	OpenOpen	ProvenProven ⬇ 2k/mo HF · 4 rankingsModerately usedModerately used	€0.45$0.50 / €1.34$1.50	675B / 41B	445 GB approx. ▣ 1 node 8 GPU disk 682 GB 1 node 8× H200/B200 (FP8) · 1 node H100/A100 (NVFP4)	Apache 2.0
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 2k HF downloads/mo · on OpenRouter · 4 leaderboards · from 2025-12 · Adoption: Moderately used (⬇ 2k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2025-12-02; data from 2026-10-09) Memory: disk 682 GB (FP8) · Min. VRAM 445 GB approx. · fits on: 1 node 8× H200/B200 (FP8) · 1 node H100/A100 (NVFP4) Official FP8 682 GB Official NVFP4 403 GB GGUF Q4_K_M 407 GB Requirement: Model card: FP8 on a B200 or H200 node; NVFP4 on an H100 or A100 node. source License: Apache 2.0 Source: artificialanalysis.ai/leaderboards/models · 2026-10-08
Mellum2.1 JetBrains	n/d	self-host	OpenOpen	Barely testedBarely t. ⬇ 110/mo HF · 0 rankingsLittle usedLittle used	n/d	12B / 2.5B	12.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 24.3 GB 1× 16 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. No independent benchmark · 110 HF downloads/mo · 19 days · Adoption: Little used (⬇ 110/mo HF · 0 rankings) (released: 2026-09-20; data from 2026-10-09) Score: New. No AA index or API price; for local coding agents. Memory: disk 24.3 GB (BF16) · Min. VRAM 12.0 GB approx. · fits on: 1× 16 GB (approx.) Official BF16 24.3 GB GGUF Q8_0 12.9 GB GGUF Q4_K_M 8.8 GB License: Apache 2.0
Liquid AI d1-3B Liquid AI	n/d	self-host	OpenOpen	NewNew ⬇ 5.4k/mo HF · 1 rankingModerately usedModerately used	n/d	3B / 3B (dense)	4.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 6.2 GB Any 8 GB GPU, Jetson or Mac	LFM Open License 1.0
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Released 4 days ago · 5k HF downloads/mo · 1 leaderboard (Decision Index 0.3) · Adoption: Moderately used (⬇ 5.4k/mo HF · 1 ranking) (leaderboards: Decision Index 0.3; released: 2026-10-05; data from 2026-10-09) Score: New (Oct 7). Decision model (not chat): 8–50 ms per answer. Memory: disk 6.2 GB (BF16) · Min. VRAM 4.0 GB approx. · fits on: Any 8 GB GPU, Jetson or Mac Official BF16 6.2 GB GGUF Q8_0 2.9 GB GGUF Q4_K_M 1.7 GB Requirement: Model card: 8 ms per decision on an RTX 4090 (bf16); also Jetson Orin Nano (50 ms) and Apple M5 Pro (30 ms). source License: LFM Open License 1.0

16 open-source models and variants in coding · 4 fit on your computer

Terminal-Bench 4.0 vs cost per task (open only)

Up = better · left = cheaper · the dashed line is the efficient frontier.

14 models · 3 on the frontier · hover over or tap a point to see its name and values

ClosedOpen weightsEfficient frontier

Swipe to see the whole chart

Price with the cheapest API that offers it; on your computer the cost is your hardware and electricity.

Table: Coding (open only)

16 models and variants · click ▸ on a row to see sources, memory and notes

Sort

Model	TB 4.0% solved ⇅	Costper task ⇅	WeightsWeights⇅	MaturityMad.⇅	€/1M$/1Min / out	AgentsAA	Paramstotal / act.	Memory ⇅VRAM · disco	License
GLM-5.3 (max) Z AI	41.9	€7.20$8.06	OpenOpen	ProvenProven OpenRouter top 20 · #4 · 5 rankingsWidely usedWidely used	€1.25$1.40 / €3.93$4.40	53.6 Opencode	753B / 40B	833 GB approx. ▣ 1 node 8 GPU disk 756 GB 1 node 8× H200 (FP8)	GLM-5.3 (condition >$10,000M MaaS)
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.5 M HF downloads/mo · top 20 OpenRouter · 5 leaderboards · 52 days · Adoption: Widely used (OpenRouter top 20 · #4 · 5 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, AA Coding Agents, LMArena Text, LMArena WebDev; released: 2026-08-18; data from 2026-10-09) Memory: disk 756 GB (FP8) · Min. VRAM 833 GB approx. · fits on: 1 node 8× H200 (FP8) Official FP8 756 GB Requirement: GenAI Brief: ~750 GB in FP8; fits on an 8× H200 node. source Repo zai-org/GLM-5.3-FP8 returns 401 (not accessible). AA Coding Agents Index: 53.6 with “Opencode - GLM-5.3” at €3.78$4.24/task License: GLM-5.3 (condition >$10,000M MaaS) Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
MiMo-V2.6-Pro Xiaomi	34.8	€0.39$0.44	OpenOpen	NewNew ⬇ 93k/mo HF · 4 rankingsModerately usedModerately used	€0.39$0.43 / €0.78$0.87	n/d	1.0T / 42B	625 GB approx. ▣ 1 node 8 GPU disk 566 GB 1 node 8× Blackwell (2 nodes on Hopper)	MIT
Featured: ★ Best open value · on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Released 18 days ago · 93k HF downloads/mo · on OpenRouter · 4 leaderboards · Adoption: Moderately used (⬇ 93k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-21; data from 2026-10-09) Memory: disk 566 GB (FP4/FP8 mixed) · Min. VRAM 625 GB approx. · fits on: 1 node 8× Blackwell (2 nodes on Hopper) Official FP4/FP8 566 GB Requirement: Model card: vLLM with tensor-parallel 8 (SGLang tp 16). GenAI Brief: ~1 TB in FP8; one node of 8 Blackwell GPUs. source License: MIT Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
GLM-5.3-Flash Z AI	32.8	€0.70$0.79	OpenOpen	ProvenProven OpenRouter top 20 · #11 · 4 rankingsWidely usedWidely used	€0.13$0.15 / €0.45$0.50	n/d	320B / 18B	363 GB approx. ▣ 1 node 8 GPU disk 328 GB 1 node 8× H100 (FP8)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 6.4 M HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 44 days · Adoption: Widely used (OpenRouter top 20 · #11 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-26; data from 2026-10-09) Memory: disk 328 GB (FP8) · Min. VRAM 363 GB approx. · fits on: 1 node 8× H100 (FP8) Official FP8 328 GB GGUF Q8_0 341 GB GGUF UD-IQ1_S 93.1 GB Requirement: GenAI Brief: Flash fits comfortably on 8× H100. source License: MIT Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
DeepSeek V4.1 Flash (max) DeepSeek	26.8	€1.03$1.16	OpenOpen	ProvenProven OpenRouter top 20 · #3 · 4 rankingsWidely usedWidely used	€0.27$0.30 / €1.07$1.20	n/d	552B / 16B	563 GB approx. ▣ 1 node 8 GPU disk 510 GB 1 node 8 GPU (FP8)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.3 M HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 29 days · Adoption: Widely used (OpenRouter top 20 · #3 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-10; data from 2026-10-09) Memory: disk 510 GB (FP8/INT8) · Min. VRAM 563 GB approx. · fits on: 1 node 8 GPU (FP8) Official FP8 510 GB GGUF Q2 (antirez) 366 GB Requirement: GenAI Brief: ~550 GB in FP8; one 8-GPU node. source License: MIT Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Qwen3.8-Flash-Next Alibaba	25.3	€0.78$0.88	OpenOpen	ProvenProven ⬇ 1.6M/mo HF · 3 rankingsWidely usedWidely used	€0.13$0.15 / €0.42$0.47	n/d	180B / 6B	124 GB approx. ▣ 2–4 GPU disk 360 GB 1× H200 (141 GB) or 2× 80 GB (approx.)	Qwen Community 1.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.6 M HF downloads/mo · 3 leaderboards · 44 days · Adoption: Widely used (⬇ 1.6M/mo HF · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena WebDev; released: 2026-08-26; data from 2026-10-09) Memory: disk 360 GB (BF16) · Min. VRAM 124 GB approx. · fits on: 1× H200 (141 GB) or 2× 80 GB (approx.) Official BF16 360 GB GGUF UD-Q4_K_XL 111 GB GGUF UD-IQ1_S 72.5 GB 180B total / 6B active: in Q4 it runs on 2× 80 GB or a Mac with 128 GB+ (approx.). License: Qwen Community 1.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
MiMo-V2.6-Flash Xiaomi	22.7	€0.21$0.23	OpenOpen	ProvenProven OpenRouter top 20 · #8 · 4 rankingsWidely usedWidely used	€0.12$0.14 / €0.25$0.28	n/d	309B / 15B	186 GB approx. ▣ 2–4 GPU disk 173 GB 4× GPU 80 GB (TP 4)	MIT
Featured: ★ Ultra-cheap open · on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 56k HF downloads/mo · top 20 OpenRouter · 4 leaderboards · 18 days · Adoption: Widely used (OpenRouter top 20 · #8 · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-09-21; data from 2026-10-09) Memory: disk 173 GB (FP4/FP8 mixed) · Min. VRAM 186 GB approx. · fits on: 4× GPU 80 GB (TP 4) Official FP4/FP8 173 GB GGUF MXFP4 167 GB Requirement: Model card: vLLM example with tensor-parallel 4. source License: MIT Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Kimi K3 (max) Moonshot AI	12.6	€4.54$5.09	OpenOpen	ProvenProven OpenRouter top 20 · #6 · 5 rankingsWidely usedWidely used	€2.68$3.00 / €13.39$15.00	51.9 Kimi Code CLI	2.8T / 104B	1719 GB approx. ▣ Multi-node disk 1561 GB Multi-node on Hopper; high-memory Blackwell node	Kimi K3 (attribution and MaaS agreement >$20M)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.2 M HF downloads/mo · top 20 OpenRouter · 5 leaderboards · 85 days · Adoption: Widely used (OpenRouter top 20 · #6 · 5 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, AA Coding Agents, LMArena Text, LMArena WebDev; released: 2026-07-16; data from 2026-10-09) Memory: disk 1561 GB (MXFP4/U8) · Min. VRAM 1719 GB approx. · fits on: Multi-node on Hopper; high-memory Blackwell node Official MXFP4 1561 GB GGUF UD-IQ1_S (community) 594 GB Requirement: GenAI Brief: ~1.4 TB in MXFP4; multi-node on Hopper. source AA Coding Agents Index: 51.9 with “Kimi Code CLI - Kimi K3” at €4.50$5.05/task License: Kimi K3 (attribution and MaaS agreement >$20M) Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Qwen3.8 2.4T A95B Alibaba	11.1	€10.25$11.49	OpenOpen	ProvenProven ⬇ 33k/mo HF · 2 rankingsWidely usedWidely used	€1.78$2.00 / €5.35$6.00	n/d	2.4T / 95B	5384 GB approx. ▣ Multi-node disk 4892 GB Multi-node (does not fit on a Hopper node)	Qwen3.8-Max (commercial terms)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 33k HF downloads/mo · on OpenRouter · 2 leaderboards (AA Intelligence Index, AA Terminal-Bench 4.0) · 58 days · Adoption: Widely used (⬇ 33k/mo HF · 2 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0; released: 2026-08-12; data from 2026-10-09) Memory: disk 4892 GB (BF16) · Min. VRAM 5384 GB approx. · fits on: Multi-node (does not fit on a Hopper node) Official BF16 4892 GB Requirement: GenAI Brief: does not fit on a Hopper node at any published precision. source License: Qwen3.8-Max (commercial terms) Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Qwen3.8 27B (xhigh) Alibaba	5.6	€3.52$3.94	OpenOpen	ProvenProven ⬇ 6.8M/mo HF · 4 rankingsWidely usedWidely used	€0.45$0.50 / €2.68$3.00	n/d	27B / 27B (dense)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 55.6 GB 1× 80 GB (BF16) · 1× 24–32 GB (Q4)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 6.8 M HF downloads/mo · on OpenRouter · 4 leaderboards · 56 days · Adoption: Widely used (⬇ 6.8M/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-14; data from 2026-10-09) Memory: disk 55.6 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 80 GB (BF16) · 1× 24–32 GB (Q4) Official BF16 55.6 GB Official FP8 30.9 GB GGUF UD-Q4_K_M 16.5 GB Requirement: GenAI Brief: 55.6 GB BF16 fits on an 80 GB GPU without quantizing; the default «single-GPU» model. source License: Apache 2.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
MiniMax-M3 MiniMax	2.0	€2.71$3.04	OpenOpen	ProvenProven ⬇ 204k/mo HF · 4 rankingsWidely usedWidely used	€0.27$0.30 / €1.07$1.20	n/d	428B / 23B	284 GB approx. ▣ 1 node 8 GPU disk 854 GB 1 node 8× H100 (8 bits)	MiniMax Community (commercial use with notice)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 204k HF downloads/mo · on OpenRouter · 4 leaderboards · from 2026-06 · Adoption: Widely used (⬇ 204k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-06-01; data from 2026-10-09) Memory: disk 854 GB (BF16) · Min. VRAM 284 GB approx. · fits on: 1 node 8× H100 (8 bits) Official BF16 854 GB Official MXFP8 444 GB GGUF MXFP4_MOE 256 GB Requirement: GenAI Brief: at 8 bits the weights fit on an 8× H100 node. source License: MiniMax Community (commercial use with notice) Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Nemotron 3 Ultra NVIDIA	0.5	€2.73$3.06	OpenOpen	ProvenProven OpenRouter top 20 · #14 · 3 rankingsWidely usedWidely used	€0.54$0.60 / €2.23$2.50	n/d	550B / 55B	390 GB approx. ▣ 1 node 8 GPU disk 1121 GB 4× B200/GB200 (NVFP4) or 8× H100	OpenMDW-1.1
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 760k HF downloads/mo · top 20 OpenRouter · 3 leaderboards · from 2026-06 · Adoption: Widely used (OpenRouter top 20 · #14 · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text; released: 2026-06-04; data from 2026-10-09) Memory: disk 1121 GB (BF16) · Min. VRAM 390 GB approx. · fits on: 4× B200/GB200 (NVFP4) or 8× H100 Official BF16 1121 GB Official NVFP4 352 GB Requirement: NVFP4 model card: minimum GPU 4× GB200, 4× B200, 4× GB300, 4× B300 or 8× H100. source License: OpenMDW-1.1 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Muse Glimmer (high) Meta	0.5	€0.24$0.27	OpenOpen	ProvenProven ⬇ 269k/mo HF · 4 rankingsWidely usedWidely used	€0.29$0.32 / €1.20$1.35	n/d	30B / n/d	19.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 59.6 GB 1× 24–32 GB (quantized)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 269k HF downloads/mo · on OpenRouter · 4 leaderboards · 60 days · Adoption: Widely used (⬇ 269k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-08-10; data from 2026-10-09) Memory: disk 59.6 GB (BF16) · Min. VRAM 19.0 GB approx. · fits on: 1× 24–32 GB (quantized) Official BF16 59.6 GB GGUF Q8_0 29.6 GB GGUF UD-Q4_K_XL 15.9 GB Requirement: Meta (via GenAI Brief): runs on 24 or 32 GB consumer GPUs. source License: Apache 2.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Gemma 4 31B Google	0.0	self-host	OpenOpen	ProvenProven ⬇ 9.6M/mo HF · 4 rankingsWidely usedWidely used	n/d	n/d	30.7B / 30.7B (dense)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 62.5 GB 1× 24 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 9.6 M HF downloads/mo · on OpenRouter · 4 leaderboards · from 2026-04 · Adoption: Widely used (⬇ 9.6M/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2026-04-02; data from 2026-10-09) Score: 0 tasks solved in Terminal-Bench 4.0. Memory: disk 62.5 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Official BF16 62.5 GB GGUF Q8_0 33.2 GB GGUF Q4_K_M 18.3 GB Requirement: Google: 12B/26B A4B/31B designed for consumer GPUs and workstations. source License: Apache 2.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
gpt-oss-120b (high) OpenAI	0.0	€0.057$0.064	OpenOpen	ProvenProven ⬇ 4.0M/mo HF · 3 rankingsWidely usedWidely used	€0.13$0.15 / €0.53$0.59	n/d	117B / 5.1B	74.0 GB approx. ▣ 1× 80 GB disk 65.2 GB 1× 80 GB (H100 / MI300X)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 4.0 M HF downloads/mo · on OpenRouter · 3 leaderboards · from 2025-08 · Adoption: Widely used (⬇ 4.0M/mo HF · 3 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text; released: 2025-08-05; data from 2026-10-09) Score: 0 tasks solved in Terminal-Bench 4.0. Memory: disk 65.2 GB (MXFP4) · Min. VRAM 74.0 GB approx. · fits on: 1× 80 GB (H100 / MI300X) Official MXFP4 65.2 GB Requirement: Model card: fits on a single 80 GB GPU (H100 or MI300X) thanks to MXFP4. source License: Apache 2.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Mistral Large 3 Mistral	0.0	€0.12$0.13	OpenOpen	ProvenProven ⬇ 2k/mo HF · 4 rankingsModerately usedModerately used	€0.45$0.50 / €1.34$1.50	n/d	675B / 41B	445 GB approx. ▣ 1 node 8 GPU disk 682 GB 1 node 8× H200/B200 (FP8) · 1 node H100/A100 (NVFP4)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 2k HF downloads/mo · on OpenRouter · 4 leaderboards · from 2025-12 · Adoption: Moderately used (⬇ 2k/mo HF · 4 rankings) (leaderboards: AA Intelligence Index, AA Terminal-Bench 4.0, LMArena Text, LMArena WebDev; released: 2025-12-02; data from 2026-10-09) Score: 0 tasks solved in Terminal-Bench 4.0. Memory: disk 682 GB (FP8) · Min. VRAM 445 GB approx. · fits on: 1 node 8× H200/B200 (FP8) · 1 node H100/A100 (NVFP4) Official FP8 682 GB Official NVFP4 403 GB GGUF Q4_K_M 407 GB Requirement: Model card: FP8 on a B200 or H200 node; NVFP4 on an H100 or A100 node. source License: Apache 2.0 Source: artificialanalysis.ai/evaluations/terminalbench-4-0 · 2026-10-08
Mellum2.1 JetBrains	n/d	self-host	OpenOpen	Barely testedBarely t. ⬇ 110/mo HF · 0 rankingsLittle usedLittle used	n/d	n/d	12B / 2.5B	12.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 24.3 GB 1× 16 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. No independent benchmark · 110 HF downloads/mo · 19 days · Adoption: Little used (⬇ 110/mo HF · 0 rankings) (released: 2026-09-20; data from 2026-10-09) Score: JetBrains' coding model; AA has not measured it on Terminal-Bench 4.0 (only its own SWE-bench Verified figures). Memory: disk 24.3 GB (BF16) · Min. VRAM 12.0 GB approx. · fits on: 1× 16 GB (approx.) Official BF16 24.3 GB GGUF Q8_0 12.9 GB GGUF Q4_K_M 8.8 GB License: Apache 2.0 Source: blog.jetbrains.com/ai/2026/10/mellum2-1-gets-to-work-a-fast-open-model · 2026-10-08

56 open-source models and variants in images · 33 fit on your computer

Quality (Elo) vs price per image (open only)

Up = better · left = cheaper · the dashed line is the efficient frontier.

48 models · 7 on the frontier · hover over or tap a point to see its name and values

ClosedOpen weightsEfficient frontier

Swipe to see the whole chart

Price with the cheapest API that offers it; on your computer the cost is your hardware and electricity.

Table: Images (open only)

56 models and variants · click ▸ on a row to see sources, memory and notes

Sort

Model	EloAA ⇅	€USDper image ⇅	WeightsWeights⇅	MaturityMad.⇅	Paramstotal / act.	Memory ⇅VRAM · disco	License
Qwen-Image-2.1 Alibaba	1035 #19	self-host	OpenOpen	NewNew ⬇ 117k/mo HF · 2 rankingsWidely usedWidely used	7.1B (transformer)	10.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 33.1 GB 1× 8–12 GB (approx.)	Qwen Research (non-commercial)
Featured: ★ Best open Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Released 19 days ago · 117k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · Adoption: Widely used (⬇ 117k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-09-20; data from 2026-10-09) Arena AA Text-to-Image: #19 · Elo 1035 (±9, 95% CI) · 5489 votes Score: Best open weights. No API (downloadable). Memory: disk 33.1 GB (BF16) · Min. VRAM 10.0 GB approx. · fits on: 1× 8–12 GB (approx.) Transformer BF16 (7.1B) 14.2 GB Transformer FP8 7.1 GB Transformer GGUF Q4_K_M 4.2 GB Encoder Qwen3-VL 8B BF16 17.5 GB Encoder FP8 9.4 GB Encoder W4A8 6.3 GB Requirement: Model card: no VRAM figure; recommends enable_model_cpu_offload() to save memory. source Full repo 33.1 GB (encoder 17.5 + transformer 14.2 + VAE 1.4). Research/evaluation-only license: commercial use requires a separate license. License: Qwen Research (non-commercial) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ideogram 4.0 (Quality) Ideogram	1012 #38	€0.089$0.10	OpenOpen	ProvenProven ⬇ 53k/mo HF · 2 rankingsModerately usedModerately used	9.3B (transformer)	12.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 27.4 GB 1× 16 GB (approx.)	Ideogram 4 Non-Commercial
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 53k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2026-06 · Adoption: Moderately used (⬇ 53k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-06-03; data from 2026-10-09) Arena AA Text-to-Image: #38 · Elo 1012 (±7, 95% CI) · 12,370 votes Score: AA text-to-image arena: #38 of 168. Memory: disk 27.4 GB (FP8) · Min. VRAM 12.0 GB approx. · fits on: 1× 16 GB (approx.) Transformer FP8 9.3 GB Transformer incondicional FP8 9.3 GB Encoder FP8 8.8 GB Transformer NF4 5.2 GB Encoder NF4 5.5 GB Weights published in FP8 and NF4 (ideogram-ai/ideogram-4 collection); gated repo. Disk = FP8 transformer + FP8 unconditional transformer + FP8 encoder. No official VRAM figure: estimate = FP8 transformer × 1.1 + 2 GB (encoders offloaded). License: Ideogram 4 Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ideogram 4.0 Ideogram	1004 #41	€0.054$0.060	OpenOpen	NewNew ⬇ 53k/mo HF · 1 rankingModerately usedModerately used	9.3B (transformer)	12.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 27.4 GB 1× 16 GB (approx.)	Ideogram 4 Non-Commercial
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Only 1 leaderboard (AA Text-to-Image) · 53k HF downloads/mo · from 2026-06 · Adoption: Moderately used (⬇ 53k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-06-03; data from 2026-10-09) Arena AA Text-to-Image: #41 · Elo 1004 (±8, 95% CI) · 11,308 votes Score: AA text-to-image arena: #41 of 168. Memory: disk 27.4 GB (FP8) · Min. VRAM 12.0 GB approx. · fits on: 1× 16 GB (approx.) Transformer FP8 9.3 GB Transformer incondicional FP8 9.3 GB Encoder FP8 8.8 GB Transformer NF4 5.2 GB Encoder NF4 5.5 GB Weights published in FP8 and NF4 (ideogram-ai/ideogram-4 collection); gated repo. Disk = FP8 transformer + FP8 unconditional transformer + FP8 encoder. No official VRAM figure: estimate = FP8 transformer × 1.1 + 2 GB (encoders offloaded). License: Ideogram 4 Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [dev] Black Forest Labs	1000 #43	€0.011$0.012	OpenOpen	ProvenProven ⬇ 408k/mo HF · 1 rankingWidely usedWidely used	32.2B (transformer)	73.0 GB approx. ▣ 1× 80 GB disk 113 GB 1× 80 GB (approx.)	FLUX Non-Commercial
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 408k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-11 · Adoption: Widely used (⬇ 408k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-11-25; data from 2026-10-09) Arena AA Text-to-Image: #43 · Elo 1000 · 20,731 votes Score: AA text-to-image arena: #43 of 168. Memory: disk 113 GB (BF16) · Min. VRAM 73.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer BF16 64.5 GB Encoder Mistral Small 3.1 24B 48.0 GB VAE 0.3 GB Gated repo. Disk = BF16 transformer + Mistral Small 3.1 24B encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Qwen Image Max 2512 Alibaba	999 #44	€0.018$0.020	OpenOpen	ProvenProven ⬇ 45k/mo HF · 1 rankingModerately usedModerately used	20.4B (transformer)	47.0 GB approx. ▣ 1× 80 GB disk 57.7 GB 1× 80 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 45k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-12 · Adoption: Moderately used (⬇ 45k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-12-30; data from 2026-10-09) Arena AA Text-to-Image: #44 · Elo 999 (±10, 95% CI) · 2186 votes Score: AA text-to-image arena: #44 of 168. Memory: disk 57.7 GB (BF16) · Min. VRAM 47.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer BF16 40.9 GB Encoder Qwen2.5-VL 7B 16.6 GB VAE 0.2 GB AA links to the Qwen-Image-2512 Space; weights at Qwen/Qwen-Image-2512. Disk = BF16 transformer + Qwen2.5-VL 7B encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [dev] Turbo Fal	998 #47	€0.0071$0.008	OpenOpen	ProvenProven ⬇ 13k/mo HF · 1 rankingModerately usedModerately used	32.2B (FLUX.2 [dev]) + LoRA	73.0 GB approx. ▣ 1× 80 GB disk 116 GB 1× 80 GB (approx.)	FLUX [dev] Non-Commercial
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 13k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-12 · Adoption: Moderately used (⬇ 13k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-12-21; data from 2026-10-09) Arena AA Text-to-Image: #47 · Elo 998 (±8, 95% CI) · 6703 votes Score: AA text-to-image arena: #47 of 168. Memory: disk 116 GB (BF16) · Min. VRAM 73.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer FLUX.2 [dev] BF16 64.5 GB LoRA Turbo (fal) 2.8 GB Encoder 48.0 GB VAE 0.3 GB It's a 2.8 GB LoRA on FLUX.2 [dev]: it needs the FLUX.2 [dev] weights. Disk = FLUX.2 [dev] BF16 transformer + Turbo LoRA (fal) + encoder + VAE. No official VRAM figure: estimate = FLUX.2 [dev] BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX [dev] Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ming-Image-0.1-Design InclusionAI	998 #48	€0.027$0.030	OpenOpen	Barely testedBarely t. ⬇ 0/mo HF · 1 rankingModerately usedModerately used	6B (according to the model card; 6.2B transformer on HF)	80.0 GB official ▣ 1× 80 GB disk 52.7 GB 1× 80 GB (validated configuration)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 0 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · 17 days · Adoption: Moderately used (⬇ 0/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-09-22; data from 2026-10-09) Arena AA Text-to-Image: #48 · Elo 998 (±7, 95% CI) · 21,708 votes Score: AA text-to-image arena: #48 of 168. Memory: disk 52.7 GB (BF16) · Min. VRAM 80.0 GB official · fits on: 1× 80 GB (validated configuration) Transformer BF16 12.3 GB MLLM 34.0 GB Connector 6.2 GB VAE 0.2 GB Requirement: Model card: «one CUDA GPU with 80 GiB VRAM (validated configuration)». source Disco = Transformer BF16 + MLLM + Connector + VAE. License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HunyuanImage 3.0 Instruct Tencent	996 #49	self-host	OpenOpen	ProvenProven ⬇ 19k/mo HF · 1 rankingWidely usedWidely used	80B / 13B	240 GB official ▣ 2–4 GPU disk 169 GB ≥ 3× 80 GB	Tencent Hunyuan Community (excludes EU, UK and South Korea)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 19k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Widely used (⬇ 19k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-26; data from 2026-10-09) Arena AA Text-to-Image: #49 · Elo 996 (±9, 95% CI) · 5602 votes Score: AA text-to-image arena: #49 of 168. No API (downloadable). Memory: disk 169 GB (BF16) · Min. VRAM 240 GB official · fits on: ≥ 3× 80 GB BF16 model 169 GB Requirement: Model card: Recommended VRAM ≥ 3 × 80 GB; 80B total (13B active). source Disk = BF16 model. License: Tencent Hunyuan Community (excludes EU, UK and South Korea) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Cosmos3-Super-Text2Image (agentic) NVIDIA	994 #50	self-host	OpenOpen	NewNew ⬇ 4.9k/mo HF · 2 rankingsModerately usedModerately used	64B	143 GB approx. ▣ 1 node 8 GPU disk 132 GB 8× H100 (recommended setup)	OpenMDW 1.1
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Limited use · 5k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2026-05 · Adoption: Moderately used (⬇ 4.9k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-05-31; data from 2026-10-09) Arena AA Text-to-Image: #50 · Elo 994 (±8, 95% CI) · 12,514 votes Score: AA text-to-image arena: #50 of 168. No API (downloadable). Memory: disk 132 GB (BF16) · Min. VRAM 143 GB approx. · fits on: 8× H100 (recommended setup) Transformer BF16 128 GB VAE 1.4 GB Vision encoder 1.2 GB Sound tokenizer 1.9 GB Requirement: Model card: recommended vLLM-Omni setup on an 8×H100 node (with --enable-layerwise-offload it fits in less memory, slower). source Disk = BF16 transformer + VAE + vision encoder + sound tokenizer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: OpenMDW 1.1 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [dev] Flash Fal	987 #55	€0.0045$0.005	OpenOpen	ProvenProven ⬇ 13k/mo HF · 1 rankingModerately usedModerately used	32.2B (FLUX.2 [dev]) + LoRA	73.0 GB approx. ▣ 1× 80 GB disk 116 GB 1× 80 GB (approx.)	FLUX [dev] Non-Commercial
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 13k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-12 · Adoption: Moderately used (⬇ 13k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-12-21; data from 2026-10-09) Arena AA Text-to-Image: #55 · Elo 987 (±9, 95% CI) · 5508 votes Score: AA text-to-image arena: #55 of 168. Memory: disk 116 GB (BF16) · Min. VRAM 73.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer FLUX.2 [dev] BF16 64.5 GB LoRA Turbo (fal) 2.8 GB Encoder 48.0 GB VAE 0.3 GB It's a 2.8 GB LoRA on FLUX.2 [dev]: it needs the FLUX.2 [dev] weights. Disk = FLUX.2 [dev] BF16 transformer + Turbo LoRA (fal) + encoder + VAE. No official VRAM figure: estimate = FLUX.2 [dev] BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX [dev] Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ideogram 4.0 Fast (Quality) Fal	986 #56	€0.016$0.018	OpenOpen	Barely testedBarely t. ⬇ 0/mo HF · 1 rankingLittle usedLittle used	9.3B (transformer)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 18.6 GB 1× 24 GB (approx.)	Ideogram 4 Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 0 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · 88 days · Adoption: Little used (⬇ 0/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-07-13; data from 2026-10-09) Arena AA Text-to-Image: #56 · Elo 986 (±9, 95% CI) · 6360 votes Score: AA text-to-image arena: #56 of 168. Memory: disk 18.6 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 18.6 GB Encoder FP8 (Ideogram 4) 8.8 GB fal repo with the distilled transformer; the encoder comes from the Ideogram 4 repo. Gated. Disk = BF16 transformer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Ideogram 4 Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Cosmos3-Super-Text2Image NVIDIA	986 #57	€0.036$0.040	OpenOpen	ProvenProven ⬇ 4.9k/mo HF · 2 rankingsModerately usedModerately used	64B	143 GB approx. ▣ 1 node 8 GPU disk 132 GB 8× H100 (recommended setup)	OpenMDW 1.1
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 5k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2026-05 · Adoption: Moderately used (⬇ 4.9k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-05-31; data from 2026-10-09) Arena AA Text-to-Image: #57 · Elo 986 (±8, 95% CI) · 11,905 votes Score: AA text-to-image arena: #57 of 168. Memory: disk 132 GB (BF16) · Min. VRAM 143 GB approx. · fits on: 8× H100 (recommended setup) Transformer BF16 128 GB VAE 1.4 GB Vision encoder 1.2 GB Sound tokenizer 1.9 GB Requirement: Model card: recommended vLLM-Omni setup on an 8×H100 node (with --enable-layerwise-offload it fits in less memory, slower). source Disk = BF16 transformer + VAE + vision encoder + sound tokenizer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: OpenMDW 1.1 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ideogram 4.0 Instant Fal	983 #65	€0.0067$0.0075	OpenOpen	Barely testedBarely t. ⬇ 0/mo HF · 1 rankingLittle usedLittle used	9.3B (transformer)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 18.6 GB 1× 24 GB (approx.)	Ideogram 4 Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 0 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · 88 days · Adoption: Little used (⬇ 0/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-07-13; data from 2026-10-09) Arena AA Text-to-Image: #65 · Elo 983 (±9, 95% CI) · 5559 votes Score: AA text-to-image arena: #65 of 168. Memory: disk 18.6 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 18.6 GB Encoder FP8 (Ideogram 4) 8.8 GB fal repo with the distilled transformer; the encoder comes from the Ideogram 4 repo. Gated. Disk = BF16 transformer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Ideogram 4 Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HiDream-O1-Image HiDream	981 #66	€0.036$0.040	OpenOpen	ProvenProven ⬇ 4.6k/mo HF · 2 rankingsModerately usedModerately used	8B	41.0 GB approx. ▣ 1× 32–48 GB disk 35.2 GB 1× 48 GB (approx.)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 5k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2026-05 · Adoption: Moderately used (⬇ 4.6k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-05-08; data from 2026-10-09) Arena AA Text-to-Image: #66 · Elo 981 (±8, 95% CI) · 11,751 votes Score: AA text-to-image arena: #66 of 168. Memory: disk 35.2 GB (F32) · Min. VRAM 41.0 GB approx. · fits on: 1× 48 GB (approx.) F32 model 35.2 GB Weights in F32 (35 GB); in BF16 they'd be half. Disk = F32 model. No official VRAM figure: estimate = F32 model × 1.1 + 2 GB (encoders offloaded). License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Cosmos3-Super-Text2Image-4Step NVIDIA	971 #69	self-host	OpenOpen	NewNew ⬇ 4.9k/mo HF · 1 rankingModerately usedModerately used	64B	143 GB approx. ▣ 1 node 8 GPU disk 132 GB 8× H100 (recommended setup)	OpenMDW 1.1
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Only 1 leaderboard (AA Text-to-Image) · 5k HF downloads/mo · 81 days · Adoption: Moderately used (⬇ 4.9k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-07-20; data from 2026-10-09) Arena AA Text-to-Image: #69 · Elo 971 (±8, 95% CI) · 10,306 votes Score: AA text-to-image arena: #69 of 168. No API (downloadable). Memory: disk 132 GB (BF16) · Min. VRAM 143 GB approx. · fits on: 8× H100 (recommended setup) Transformer BF16 128 GB VAE 1.4 GB Vision encoder 1.2 GB Sound tokenizer 1.9 GB Requirement: Model card: recommended vLLM-Omni setup on an 8×H100 node (with --enable-layerwise-offload it fits in less memory, slower). source Disk = BF16 transformer + VAE + vision encoder + sound tokenizer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: OpenMDW 1.1 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Ideogram 4.0 Fast Fal	969 #71	€0.0094$0.011	OpenOpen	Barely testedBarely t. ⬇ 0/mo HF · 1 rankingLittle usedLittle used	9.3B (transformer)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 18.6 GB 1× 24 GB (approx.)	Ideogram 4 Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 0 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · 88 days · Adoption: Little used (⬇ 0/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-07-13; data from 2026-10-09) Arena AA Text-to-Image: #71 · Elo 969 (±9, 95% CI) · 5490 votes Score: AA text-to-image arena: #71 of 168. Memory: disk 18.6 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 18.6 GB Encoder FP8 (Ideogram 4) 8.8 GB fal repo with the distilled transformer; the encoder comes from the Ideogram 4 repo. Gated. Disk = BF16 transformer. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Ideogram 4 Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HunyuanImage 3.0 Tencent	945 #79	self-host	OpenOpen	NewNew ⬇ 3.4k/mo HF · 2 rankingsWidely usedWidely used	80B / 13B	240 GB official ▣ 2–4 GPU disk 169 GB ≥ 3× 80 GB	Tencent Hunyuan Community (excludes EU, UK and South Korea)
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Limited use · 3k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2025-09 · Adoption: Widely used (⬇ 3.4k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2025-09-28; data from 2026-10-09) Arena AA Text-to-Image: #79 · Elo 945 (±11, 95% CI) · 2159 votes Score: AA text-to-image arena: #79 of 168. No API (downloadable). Memory: disk 169 GB (BF16) · Min. VRAM 240 GB official · fits on: ≥ 3× 80 GB BF16 model 169 GB Requirement: Model card (HunyuanImage-3.0-Instruct table): Recommended VRAM ≥ 3 × 80 GB; 80B total (13B active). source Disk = BF16 model. License: Tencent Hunyuan Community (excludes EU, UK and South Korea) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [klein] 9B Black Forest Labs	941 #82	€0.013$0.015	OpenOpen	ProvenProven ⬇ 194k/mo HF · 1 rankingWidely usedWidely used	9.1B (transformer)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 34.7 GB 1× 24 GB (approx.)	FLUX Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 194k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Widely used (⬇ 194k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-15; data from 2026-10-09) Arena AA Text-to-Image: #82 · Elo 941 (±8, 95% CI) · 11,820 votes Score: AA text-to-image arena: #82 of 168. Memory: disk 34.7 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 18.2 GB Encoder 16.4 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Z-Image Turbo Alibaba	941 #83	€0.0045$0.005	OpenOpen	ProvenProven ⬇ 623k/mo HF · 2 rankingsWidely usedWidely used	6.2B (transformer)	16.0 GB official ▣ 1 consumer GPU ✓ Fits on your computer disk 32.8 GB 1× 16 GB (consumer)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 623k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2025-12 · Adoption: Widely used (⬇ 623k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2025-12-02; data from 2026-10-09) Arena AA Text-to-Image: #83 · Elo 941 (±8, 95% CI) · 8880 votes Score: AA text-to-image arena: #83 of 168. Memory: disk 32.8 GB (F32) · Min. VRAM 16.0 GB official · fits on: 1× 16 GB (consumer) Transformer F32 24.6 GB Encoder Qwen3 4B 8.0 GB VAE 0.2 GB Requirement: Model card: «fits comfortably within 16G VRAM consumer devices». source Disco = Transformer F32 + Encoder Qwen3 4B + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
ERNIE Image Turbo Baidu	924 #87	€0.0094$0.011	OpenOpen	NewNew ⬇ 3.6k/mo HF · 1 rankingModerately usedModerately used	8B (transformer)	24.0 GB official ▣ 1 consumer GPU ✓ Fits on your computer disk 31.6 GB 1× 24 GB (consumer)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Only 1 leaderboard (AA Text-to-Image) · 4k HF downloads/mo · from 2026-04 · Adoption: Moderately used (⬇ 3.6k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-04-15; data from 2026-10-09) Arena AA Text-to-Image: #87 · Elo 924 (±8, 95% CI) · 11,916 votes Score: AA text-to-image arena: #87 of 168. Memory: disk 31.6 GB (BF16) · Min. VRAM 24.0 GB official · fits on: 1× 24 GB (consumer) Transformer BF16 16.1 GB Encoder 7.7 GB Prompt enhancer 7.7 GB VAE 0.2 GB Requirement: Model card: «can run on consumer GPUs with 24G VRAM». source Disco = Transformer BF16 + Encoder + Prompt enhancer + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
ERNIE Image Baidu	912 #97	€0.028$0.032	OpenOpen	NewNew ⬇ 1.2k/mo HF · 1 rankingModerately usedModerately used	8B (transformer)	24.0 GB official ▣ 1 consumer GPU ✓ Fits on your computer disk 31.6 GB 1× 24 GB (consumer)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Only 1 leaderboard (AA Text-to-Image) · 1k HF downloads/mo · from 2026-04 · Adoption: Moderately used (⬇ 1.2k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-04-15; data from 2026-10-09) Arena AA Text-to-Image: #97 · Elo 912 (±8, 95% CI) · 12,180 votes Score: AA text-to-image arena: #97 of 168. Memory: disk 31.6 GB (BF16) · Min. VRAM 24.0 GB official · fits on: 1× 24 GB (consumer) Transformer BF16 16.1 GB Encoder 7.7 GB Prompt enhancer 7.7 GB VAE 0.2 GB Requirement: Model card: «can run on consumer GPUs with 24G VRAM». source Disco = Transformer BF16 + Encoder + Prompt enhancer + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [klein] Base 9B Black Forest Labs	903 #100	€0.0098$0.011	OpenOpen	ProvenProven ⬇ 93k/mo HF · 1 rankingModerately usedModerately used	9.1B (transformer)	22.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 34.7 GB 1× 24 GB (approx.)	FLUX Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 93k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Moderately used (⬇ 93k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-15; data from 2026-10-09) Arena AA Text-to-Image: #100 · Elo 903 (±9, 95% CI) · 2766 votes Score: AA text-to-image arena: #100 of 168. Memory: disk 34.7 GB (BF16) · Min. VRAM 22.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 18.2 GB Encoder 16.4 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
GLM-Image Zhipu (Z.ai)	891 #106	€0.045$0.050	OpenOpen	ProvenProven ⬇ 4.7k/mo HF · 2 rankingsWidely usedWidely used	6.9B (transformer) + vision-language encoder	23.0 GB official (approx.) ▣ 1 consumer GPU ✓ Fits on your computer disk 35.8 GB ~23 GB with CPU offload	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 5k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2026-01 · Adoption: Widely used (⬇ 4.7k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2026-01-13; data from 2026-10-09) Arena AA Text-to-Image: #106 · Elo 891 (±7, 95% CI) · 5345 votes Score: AA text-to-image arena: #106 of 168. Memory: disk 35.8 GB (BF16) · Min. VRAM 23.0 GB official (approx.) · fits on: ~23 GB with CPU offload Transformer BF16 13.8 GB Vision-language encoder 20.2 GB Text encoder 0.9 GB VAE 0.8 GB Requirement: Model card: with enable_model_cpu_offload=True it runs with ~23GB of GPU memory (slower). source Disk = BF16 transformer + vision-language encoder + text encoder + VAE. License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Qwen Image Alibaba	890 #107	€0.018$0.020	OpenOpen	ProvenProven ⬇ 266k/mo HF · 2 rankingsWidely usedWidely used	20.4B (transformer)	47.0 GB approx. ▣ 1× 80 GB disk 57.7 GB 1× 80 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 266k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2025-08 · Adoption: Widely used (⬇ 266k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2025-08-04; data from 2026-10-09) Arena AA Text-to-Image: #107 · Elo 890 (±7, 95% CI) · 5085 votes Score: AA text-to-image arena: #107 of 168. Memory: disk 57.7 GB (BF16) · Min. VRAM 47.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer BF16 40.9 GB Encoder Qwen2.5-VL 7B 16.6 GB VAE 0.2 GB Disk = BF16 transformer + Qwen2.5-VL 7B encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HunyuanImage 2.1 Tencent	887 #108	€0.089$0.10	OpenOpen	Barely testedBarely t. ⬇ 170/mo HF · 1 rankingModerately usedModerately used	17B (DiT)	24.0 GB official (with offload) ▣ 1 consumer GPU ✓ Fits on your computer disk 64.9 GB 24 GB (2K, FP8 + CPU offload)	Tencent Hunyuan Community (excludes EU, UK and South Korea)
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 170 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-09 · Adoption: Moderately used (⬇ 170/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-09-08; data from 2026-10-09) Arena AA Text-to-Image: #108 · Elo 887 (±9, 95% CI) · 3579 votes Score: AA text-to-image arena: #108 of 168. Memory: disk 64.9 GB (BF16) · Min. VRAM 24.0 GB official (with offload) · fits on: 24 GB (2K, FP8 + CPU offload) DiT BF16 34.9 GB DiT FP8 17.9 GB DiT destilado BF16 34.9 GB Refiner BF16 30.1 GB Reprompt 15.0 GB Requirement: Model card: minimum 24 GB of GPU memory for 2048×2048, measured with CPU offload and FP8. source Repo with several DiT variants (151 GB in total); disk = DiT + BF16 refiner. Encoders separate. Disk = BF16 DiT + BF16 refiner. License: Tencent Hunyuan Community (excludes EU, UK and South Korea) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FIBO Bria	881 #113	€0.036$0.040	OpenOpen	Barely testedBarely t. ⬇ 562/mo HF · 1 rankingModerately usedModerately used	8.3B (transformer)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 25.5 GB 1× 24 GB (approx.)	Bria FIBO (non-commercial)
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 562 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-10 · Adoption: Moderately used (⬇ 562/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-10-30; data from 2026-10-09) Arena AA Text-to-Image: #113 · Elo 881 (±7, 95% CI) · 4804 votes Score: AA text-to-image arena: #113 of 168. Memory: disk 25.5 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 16.6 GB Encoder 6.2 GB VAE 2.8 GB Gated repo. Disk = BF16 transformer + encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Bria FIBO (non-commercial) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Z-Image Base Alibaba	878 #115	€0.0089$0.010	OpenOpen	ProvenProven ⬇ 54k/mo HF · 1 rankingWidely usedWidely used	6.2B (transformer)	16.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 20.5 GB 1× 24 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 54k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Widely used (⬇ 54k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-27; data from 2026-10-09) Arena AA Text-to-Image: #115 · Elo 878 (±7, 95% CI) · 5130 votes Score: AA text-to-image arena: #115 of 168. Memory: disk 20.5 GB (BF16) · Min. VRAM 16.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 12.3 GB Encoder Qwen3 4B 8.0 GB VAE 0.2 GB AA doesn't mark it as open, but the weights are on HF (Tongyi-MAI/Z-Image, Apache 2.0). Disk = BF16 transformer + Qwen3 4B encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HiDream-O1-Image-Dev HiDream	875 #116	€0.0045$0.005	OpenOpen	Barely testedBarely t. ⬇ 657/mo HF · 1 rankingModerately usedModerately used	8B	41.0 GB approx. ▣ 1× 32–48 GB disk 35.2 GB 1× 48 GB (approx.)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 657 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-05 · Adoption: Moderately used (⬇ 657/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-05-08; data from 2026-10-09) Arena AA Text-to-Image: #116 · Elo 875 (±8, 95% CI) · 4080 votes Score: AA text-to-image arena: #116 of 168. Memory: disk 35.2 GB (F32) · Min. VRAM 41.0 GB approx. · fits on: 1× 48 GB (approx.) F32 model 35.2 GB Disk = F32 model. No official VRAM figure: estimate = F32 model × 1.1 + 2 GB (encoders offloaded). License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Infinity 8B ByteDance	875 #118	€0.0015$0.0017	OpenOpen	Barely testedBarely t. ⬇ 103/mo HF · 1 rankingLittle usedLittle used	8B	39.0 GB approx. ▣ 1× 32–48 GB disk 34.7 GB 1× 48 GB (approx.)	MIT
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 103 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-02 · Adoption: Little used (⬇ 103/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-02-18; data from 2026-10-09) Arena AA Text-to-Image: #118 · Elo 875 (±9, 95% CI) · 2406 votes Score: AA text-to-image arena: #118 of 168. Memory: disk 34.7 GB (BF16) · Min. VRAM 39.0 GB approx. · fits on: 1× 48 GB (approx.) Infinity 8B (1024px) 33.5 GB VAE d56 1.2 GB The repo also includes 2B, 125M and a 512px version (148 GB in total). Disk = Infinity 8B (1024px) + VAE d56. No official VRAM figure: estimate = Infinity 8B (1024px) × 1.1 + 2 GB (encoders offloaded). License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HiDream-I1-Dev HiDream	874 #119	€0.021$0.024	OpenOpen	Barely testedBarely t. ⬇ 292/mo HF · 1 rankingModerately usedModerately used	17.1B (transformer)	40.0 GB approx. ▣ 1× 32–48 GB disk 47.2 GB 1× 48 GB (approx.)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 292 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-04 · Adoption: Moderately used (⬇ 292/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-04-07; data from 2026-10-09) Arena AA Text-to-Image: #119 · Elo 874 (±10, 95% CI) · 1914 votes Score: AA text-to-image arena: #119 of 168. Memory: disk 47.2 GB (BF16) · Min. VRAM 40.0 GB approx. · fits on: 1× 48 GB (approx.) Transformer 34.2 GB Encoder T5-XXL 9.5 GB Encoders CLIP 12.8 GB VAE 0.2 GB Also needs Llama-3.1-8B-Instruct as the encoder (separate repo). Disk = Transformer + CLIP encoders + VAE. No official VRAM figure: estimate = Transformer × 1.1 + 2 GB (encoders offloaded). License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
HiDream-I1-Fast HiDream	870 #121	€0.011$0.012	OpenOpen	ProvenProven ⬇ 54k/mo HF · 1 rankingModerately usedModerately used	17.1B (transformer)	40.0 GB approx. ▣ 1× 32–48 GB disk 47.2 GB 1× 48 GB (approx.)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 54k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-04 · Adoption: Moderately used (⬇ 54k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-04-07; data from 2026-10-09) Arena AA Text-to-Image: #121 · Elo 870 (±9, 95% CI) · 2430 votes Score: AA text-to-image arena: #121 of 168. Memory: disk 47.2 GB (BF16) · Min. VRAM 40.0 GB approx. · fits on: 1× 48 GB (approx.) Transformer 34.2 GB Encoder T5-XXL 9.5 GB Encoders CLIP 12.8 GB VAE 0.2 GB Also needs Llama-3.1-8B-Instruct as the encoder (separate repo). Disk = Transformer + CLIP encoders + VAE. No official VRAM figure: estimate = Transformer × 1.1 + 2 GB (encoders offloaded). License: MIT Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
SRPO Tencent	864 #124	€0.023$0.026	OpenOpen	Barely testedBarely t. ⬇ 538/mo HF · 1 rankingModerately usedModerately used	11.9B (transformer, fine-tune of FLUX.1 [dev])	54.0 GB approx. ▣ 1× 80 GB disk 47.6 GB 1× 80 GB (approx.)	Tencent Hunyuan Community (according to HF; territory not verified)
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 538 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-09 · Adoption: Moderately used (⬇ 538/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-09-11; data from 2026-10-09) Arena AA Text-to-Image: #124 · Elo 864 (±7, 95% CI) · 5057 votes Score: AA text-to-image arena: #124 of 168. Memory: disk 47.6 GB (F32) · Min. VRAM 54.0 GB approx. · fits on: 1× 80 GB (approx.) Transformer F32 47.6 GB Transformer only (F32); uses the encoders and VAE of FLUX.1 [dev]. Disk = F32 transformer. No official VRAM figure: estimate = F32 transformer × 1.1 + 2 GB (encoders offloaded). License: Tencent Hunyuan Community (according to HF; territory not verified) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [klein] 4B Black Forest Labs	863 #125	€0.012$0.014	OpenOpen	ProvenProven ⬇ 404k/mo HF · 1 rankingWidely usedWidely used	3.9B (transformer)	13.0 GB official (approx.) ▣ 1 consumer GPU ✓ Fits on your computer disk 16.0 GB ~13 GB (RTX 3090/4070 or better)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 404k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Widely used (⬇ 404k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-15; data from 2026-10-09) Arena AA Text-to-Image: #125 · Elo 863 (±9, 95% CI) · 6884 votes Score: AA text-to-image arena: #125 of 168. Memory: disk 16.0 GB (BF16) · Min. VRAM 13.0 GB official (approx.) · fits on: ~13 GB (RTX 3090/4070 or better) Transformer BF16 7.8 GB Encoder 8.0 GB VAE 0.2 GB Requirement: Model card: «fits in ~13GB VRAM and is accessible on NVIDIA RTX 3090/4070 and above». source Disco = Transformer BF16 + Encoder + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
LongCat Image Meituan	861 #127	€0.12$0.13	OpenOpen	ProvenProven ⬇ 14k/mo HF · 1 rankingModerately usedModerately used	6B	17.0 GB official (approx.) ▣ 1 consumer GPU ✓ Fits on your computer disk 29.3 GB ~17 GB with CPU offload	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 14k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-12 · Adoption: Moderately used (⬇ 14k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-12-05; data from 2026-10-09) Arena AA Text-to-Image: #127 · Elo 861 (±8, 95% CI) · 3847 votes Score: AA text-to-image arena: #127 of 168. Memory: disk 29.3 GB (BF16) · Min. VRAM 17.0 GB official (approx.) · fits on: ~17 GB with CPU offload Transformer BF16 12.5 GB Encoder 16.6 GB VAE 0.2 GB Requirement: Model card: enable_model_cpu_offload() «Required ~17 GB». source Disco = Transformer BF16 + Encoder + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.1 Krea [dev] Black Forest Labs	845 #129	€0.022$0.025	OpenOpen	ProvenProven ⬇ 41k/mo HF · 1 rankingModerately usedModerately used	11.9B (transformer)	28.0 GB approx. ▣ 1 consumer GPU disk 33.7 GB 1× 32 GB (approx.)	FLUX.1 [dev] Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 41k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-07 · Adoption: Moderately used (⬇ 41k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-07-31; data from 2026-10-09) Arena AA Text-to-Image: #129 · Elo 845 (±9, 95% CI) · 3489 votes Score: AA text-to-image arena: #129 of 168. Memory: disk 33.7 GB (BF16) · Min. VRAM 28.0 GB approx. · fits on: 1× 32 GB (approx.) Transformer BF16 23.8 GB Encoder T5-XXL 9.5 GB Encoder CLIP 0.2 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + T5-XXL encoder + CLIP encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX.1 [dev] Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.1 [dev] Black Forest Labs	840 #132	€0.022$0.025	OpenOpen	ProvenProven ⬇ 692k/mo HF · 1 rankingWidely usedWidely used	11.9B (transformer)	28.0 GB approx. ▣ 1 consumer GPU disk 33.7 GB 1× 32 GB (approx.)	FLUX.1 [dev] Non-Commercial
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 692k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-08 · Adoption: Widely used (⬇ 692k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-08-01; data from 2026-10-09) Arena AA Text-to-Image: #132 · Elo 840 (±7, 95% CI) · 9780 votes Score: AA text-to-image arena: #132 of 168. Memory: disk 33.7 GB (BF16) · Min. VRAM 28.0 GB approx. · fits on: 1× 32 GB (approx.) Transformer BF16 23.8 GB Encoder T5-XXL 9.5 GB Encoder CLIP 0.2 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + T5-XXL encoder + CLIP encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: FLUX.1 [dev] Non-Commercial Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 3.5 Large Stability AI	840 #133	€0.058$0.065	OpenOpen	ProvenProven ⬇ 102k/mo HF · 1 rankingWidely usedWidely used	8.1B (transformer)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 55.1 GB 1× 24 GB (approx.)	Stability AI Community (free < $1M/year)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 102k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-10 · Adoption: Widely used (⬇ 102k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-10-22; data from 2026-10-09) Arena AA Text-to-Image: #133 · Elo 840 (±9, 95% CI) · 5129 votes Score: AA text-to-image arena: #133 of 168. Memory: disk 55.1 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 16.3 GB Encoder T5-XXL 19.1 GB Encoders CLIP 38.6 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + CLIP encoders + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Stability AI Community (free < $1M/year) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 3.5 Large Turbo Stability AI	829 #137	€0.036$0.040	OpenOpen	ProvenProven ⬇ 6.3k/mo HF · 1 rankingModerately usedModerately used	8.1B (transformer)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 55.1 GB 1× 24 GB (approx.)	Stability AI Community (free < $1M/year)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 6k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-10 · Adoption: Moderately used (⬇ 6.3k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-10-22; data from 2026-10-09) Arena AA Text-to-Image: #137 · Elo 829 (±9, 95% CI) · 5032 votes Score: AA text-to-image arena: #137 of 168. Memory: disk 55.1 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 16.3 GB Encoder T5-XXL 19.1 GB Encoders CLIP 38.6 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + CLIP encoders + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Stability AI Community (free < $1M/year) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FIBO Lite Bria	826 #138	€0.018$0.020	OpenOpen	Barely testedBarely t. ⬇ 64/mo HF · 1 rankingLittle usedLittle used	8.3B (transformer)	20.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 24.1 GB 1× 24 GB (approx.)	Bria FIBO (non-commercial)
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 64 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-11 · Adoption: Little used (⬇ 64/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-11-27; data from 2026-10-09) Arena AA Text-to-Image: #138 · Elo 826 (±11, 95% CI) · 3247 votes Score: AA text-to-image arena: #138 of 168. Memory: disk 24.1 GB (BF16) · Min. VRAM 20.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 16.6 GB Encoder 6.2 GB VAE 1.4 GB Gated repo. Disk = BF16 transformer + encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Bria FIBO (non-commercial) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.1 [schnell] Black Forest Labs	802 #143	€0.0027$0.003	OpenOpen	ProvenProven ⬇ 651k/mo HF · 1 rankingWidely usedWidely used	11.9B (transformer)	28.0 GB approx. ▣ 1 consumer GPU disk 33.7 GB 1× 32 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 651k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-08 · Adoption: Widely used (⬇ 651k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-08-02; data from 2026-10-09) Arena AA Text-to-Image: #143 · Elo 802 (±7, 95% CI) · 11,521 votes Score: AA text-to-image arena: #143 of 168. Memory: disk 33.7 GB (BF16) · Min. VRAM 28.0 GB approx. · fits on: 1× 32 GB (approx.) Transformer BF16 23.8 GB Encoder T5-XXL 9.5 GB Encoder CLIP 0.2 GB VAE 0.2 GB Disk = BF16 transformer + T5-XXL encoder + CLIP encoder + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
FLUX.2 [klein] Base 4B Black Forest Labs	794 #146	€0.008$0.009	OpenOpen	ProvenProven ⬇ 306k/mo HF · 1 rankingWidely usedWidely used	3.9B (transformer)	13.0 GB official (approx.) ▣ 1 consumer GPU ✓ Fits on your computer disk 16.0 GB ~13 GB (RTX 3090/4070 or better)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 306k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2026-01 · Adoption: Widely used (⬇ 306k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2026-01-15; data from 2026-10-09) Arena AA Text-to-Image: #146 · Elo 794 (±10, 95% CI) · 3219 votes Score: AA text-to-image arena: #146 of 168. Memory: disk 16.0 GB (BF16) · Min. VRAM 13.0 GB official (approx.) · fits on: ~13 GB (RTX 3090/4070 or better) Transformer BF16 7.8 GB Encoder 8.0 GB VAE 0.2 GB Requirement: Model card: «fits in ~13GB VRAM and is accessible on NVIDIA RTX 3090/4070 and above». source Disco = Transformer BF16 + Encoder + VAE. License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Lumina Image v2 OpenGVLab	782 #148	€0.067$0.075	OpenOpen	ProvenProven ⬇ 1.1k/mo HF · 1 rankingModerately usedModerately used	2.6B (transformer)	13.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 21.2 GB 1× 16 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 1k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-01 · Adoption: Moderately used (⬇ 1.1k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-01-25; data from 2026-10-09) Arena AA Text-to-Image: #148 · Elo 782 (±9, 95% CI) · 4438 votes Score: AA text-to-image arena: #148 of 168. Memory: disk 21.2 GB (F32) · Min. VRAM 13.0 GB approx. · fits on: 1× 16 GB (approx.) Transformer F32 10.4 GB Encoder Gemma 2 10.5 GB VAE 0.3 GB AA links to GitHub; weights at Alpha-VLLM/Lumina-Image-2.0. Disk = F32 transformer + Gemma 2 encoder + VAE. No official VRAM figure: estimate = F32 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Playground v2.5 Playground AI	777 #150	€0.071$0.080	OpenOpen	ProvenProven ⬇ 260k/mo HF · 1 rankingWidely usedWidely used	2.6B (UNet)	19.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 20.8 GB 1× 24 GB (approx.)	Playground v2.5 Community
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 260k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-02 · Adoption: Widely used (⬇ 260k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-02-27; data from 2026-10-09) Arena AA Text-to-Image: #150 · Elo 777 (±8, 95% CI) · 11,732 votes Score: AA text-to-image arena: #150 of 168. Memory: disk 20.8 GB (F32) · Min. VRAM 19.0 GB approx. · fits on: 1× 24 GB (approx.) UNet 15.4 GB Encoders CLIP 4.9 GB VAE 0.5 GB Disk = UNet + CLIP encoders + VAE. No official VRAM figure: estimate = UNet × 1.1 + 2 GB (encoders offloaded). License: Playground v2.5 Community Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 3.5 Medium Stability AI	760 #153	€0.018$0.020	OpenOpen	ProvenProven ⬇ 86k/mo HF · 1 rankingWidely usedWidely used	2.5B (transformer)	7.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 43.8 GB 1× 8–12 GB (approx.)	Stability AI Community (free < $1M/year)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 86k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-10 · Adoption: Widely used (⬇ 86k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-10-29; data from 2026-10-09) Arena AA Text-to-Image: #153 · Elo 760 (±10, 95% CI) · 2172 votes Score: AA text-to-image arena: #153 of 168. Memory: disk 43.8 GB (BF16) · Min. VRAM 7.0 GB approx. · fits on: 1× 8–12 GB (approx.) Transformer BF16 4.9 GB Encoder T5-XXL 19.1 GB Encoders CLIP 38.6 GB VAE 0.2 GB Gated repo. Disk = BF16 transformer + CLIP encoders + VAE. No official VRAM figure: estimate = BF16 transformer × 1.1 + 2 GB (encoders offloaded). License: Stability AI Community (free < $1M/year) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Sana Sprint 1.6B NVIDIA	754 #154	€0.0013$0.0015	OpenOpen	Barely testedBarely t. ⬇ 71/mo HF · 1 rankingLittle usedLittle used	1.6B	9.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 6.5 GB 1× 8–12 GB (approx.)	Apache 2.0
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Very little use · 71 HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-03 · Adoption: Little used (⬇ 71/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-03-22; data from 2026-10-09) Arena AA Text-to-Image: #154 · Elo 754 (±10, 95% CI) · 1973 votes Score: AA text-to-image arena: #154 of 168. Memory: disk 6.5 GB (F32) · Min. VRAM 9.0 GB approx. · fits on: 1× 8–12 GB (approx.) Checkpoint (.pth) 6.5 GB Separate Gemma 2 2B encoder. Disk = Checkpoint (.pth). No official VRAM figure: estimate = Checkpoint (.pth) × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
OmniGen V2 VectorSpaceLab	727 #156	€0.13$0.15	OpenOpen	ProvenProven ⬇ 3.5k/mo HF · 1 rankingModerately usedModerately used	4B (transformer) + MLLM	19.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 31.2 GB 1× 24 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 4k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-06 · Adoption: Moderately used (⬇ 3.5k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-06-16; data from 2026-10-09) Arena AA Text-to-Image: #156 · Elo 727 (±9, 95% CI) · 4462 votes Score: AA text-to-image arena: #156 of 168. Memory: disk 31.2 GB (F32) · Min. VRAM 19.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer F32 15.9 GB MLLM Qwen2.5-VL 15.0 GB VAE 0.3 GB Requirement: Model card: CPU offload cuts VRAM by almost 50%; sequential offload < 3 GB (slow). No base figure. source Disk = F32 transformer + Qwen2.5-VL MLLM + VAE. No official VRAM figure: estimate = F32 transformer × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 3 Medium Stability AI	720 #158	€0.031$0.035	OpenOpen	ProvenProven ⬇ 2.7k/mo HF · 1 rankingWidely usedWidely used	n/d	7.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 15.8 GB 1× 8–12 GB (approx.)	Stability AI Community (free < $1M/year)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 3k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-06 · Adoption: Widely used (⬇ 2.7k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-06-12; data from 2026-10-09) Arena AA Text-to-Image: #158 · Elo 720 (±9, 95% CI) · 9971 votes Score: AA text-to-image arena: #158 of 168. Memory: disk 15.8 GB (FP16) · Min. VRAM 7.0 GB approx. · fits on: 1× 8–12 GB (approx.) Checkpoint without encoders 4.3 GB With CLIP + T5 FP8 10.9 GB With CLIP + T5 FP16 15.8 GB Gated repo; HF doesn't publish the number of parameters. Disk = With CLIP + T5 FP16. No official VRAM figure: estimate = checkpoint without encoders × 1.1 + 2 GB (encoders offloaded). License: Stability AI Community (free < $1M/year) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
SDXL Lightning ByteDance	720 #159	€0.0017$0.0019	OpenOpen	ProvenProven ⬇ 108k/mo HF · 1 rankingWidely usedWidely used	2.6B (SDXL UNet)	8.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 6.9 GB 1× 8–12 GB (approx.)	CreativeML OpenRAIL++-M
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 108k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2024-02 · Adoption: Widely used (⬇ 108k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2024-02-21; data from 2026-10-09) Arena AA Text-to-Image: #159 · Elo 720 (±9, 95% CI) · 11,478 votes Score: AA text-to-image arena: #159 of 168. Memory: disk 6.9 GB (FP16) · Min. VRAM 8.0 GB approx. · fits on: 1× 8–12 GB (approx.) UNet 4 pasos 5.1 GB Checkpoint completo 4 pasos 6.9 GB LoRA 4 pasos 0.4 GB Distilled from SDXL (repo with 1, 2, 4 and 8-step variants; 49 GB in total). Disk = full 4-step checkpoint. No official VRAM figure: estimate = 4-step UNet × 1.1 + 2 GB (encoders offloaded). License: CreativeML OpenRAIL++-M Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Bagel ByteDance	706 #162	€0.089$0.10	OpenOpen	ProvenProven ⬇ 1k/mo HF · 2 rankingsWidely usedWidely used	14B / 7B	34.0 GB approx. ▣ 1× 32–48 GB disk 29.6 GB 1× 48 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1k HF downloads/mo · 2 leaderboards (AA Text-to-Image, LMArena T2I) · from 2025-05 · Adoption: Widely used (⬇ 1k/mo HF · 2 rankings) (leaderboards: AA Text-to-Image, LMArena T2I; released: 2025-05-20; data from 2026-10-09) Arena AA Text-to-Image: #162 · Elo 706 (±9, 95% CI) · 4181 votes Score: AA text-to-image arena: #162 of 168. Memory: disk 29.6 GB (BF16) · Min. VRAM 34.0 GB approx. · fits on: 1× 48 GB (approx.) Model (EMA) 29.2 GB Autoencoder 0.3 GB Disk = Model (EMA) + Autoencoder. No official VRAM figure: estimate = Model (EMA) × 1.1 + 2 GB (encoders offloaded). License: Apache 2.0 Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Bria 3.2 Bria	705 #163	€0.036$0.040	OpenOpen	ProvenProven 1 rankingNo usage dataNo data	n/d	Repo not accessible without an account (HTTP 401): size not verified.	n/a (repo not accessible)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 1 leaderboard (AA Text-to-Image) · from 2025-07 · Adoption: No usage data (1 ranking) (leaderboards: AA Text-to-Image; released: 2025-07-22; data from 2026-10-09) Arena AA Text-to-Image: #163 · Elo 705 (±10, 95% CI) · 2658 votes Score: AA text-to-image arena: #163 of 168. Memory: Repo not accessible without an account (HTTP 401): size not verified. License: n/a (repo not accessible) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion XL 1.0 Stability AI	678 #164	€0.008$0.009	OpenOpen	ProvenProven ⬇ 4.3M/mo HF · 1 rankingWidely usedWidely used	2.6B (UNet)	10.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 6.9 GB 1× 8–12 GB (approx.)	CreativeML OpenRAIL++-M
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Massive use · 4.3 M HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2023-07 · Adoption: Widely used (⬇ 4.3M/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2023-07-23; data from 2026-10-09) Arena AA Text-to-Image: #164 · Elo 678 (±9, 95% CI) · 11,633 votes Score: AA text-to-image arena: #164 of 168. Memory: disk 6.9 GB (FP16) · Min. VRAM 10.0 GB approx. · fits on: 1× 8–12 GB (approx.) Single FP16 checkpoint 6.9 GB UNet (diffusers) 10.3 GB Disk = single FP16 checkpoint. No official VRAM figure: estimate = single FP16 checkpoint × 1.1 + 2 GB (encoders offloaded). License: CreativeML OpenRAIL++-M Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 2.1 Stability AI	549 #165	€0.0035$0.0039	OpenOpen	ProvenProven 1 rankingNo usage dataNo data	n/d	Repo no longer accessible on HF (HTTP 401): size not verified.	n/a (repo not accessible)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 1 leaderboard (AA Text-to-Image) · from 2022-12 · Adoption: No usage data (1 ranking) (leaderboards: AA Text-to-Image; released: 2022-12-07; data from 2026-10-09) Arena AA Text-to-Image: #165 · Elo 549 (±11, 95% CI) · 13,596 votes Score: AA text-to-image arena: #165 of 168. Memory: Repo no longer accessible on HF (HTTP 401): size not verified. License: n/a (repo not accessible) Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Janus Pro DeepSeek	533 #167	self-host	OpenOpen	ProvenProven ⬇ 11k/mo HF · 1 rankingWidely usedWidely used	7B	18.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 14.8 GB 1× 24 GB (approx.)	MIT (code) + DeepSeek Model License
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Veteran · 11k HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2025-01 · Adoption: Widely used (⬇ 11k/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2025-01-27; data from 2026-10-09) Arena AA Text-to-Image: #167 · Elo 533 (±12, 95% CI) · 4696 votes Score: AA text-to-image arena: #167 of 168. No API (downloadable). Memory: disk 14.8 GB (BF16) · Min. VRAM 18.0 GB approx. · fits on: 1× 24 GB (approx.) Model (.bin) 14.8 GB AA links to GitHub; weights at deepseek-ai/Janus-Pro-7B (there is also a 1B). Disk = Model (.bin). No official VRAM figure: estimate = Model (.bin) × 1.1 + 2 GB (encoders offloaded). License: MIT (code) + DeepSeek Model License Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Stable Diffusion 1.5 Stability AI	456 #168	€0.039$0.044	OpenOpen	ProvenProven ⬇ 1.6M/mo HF · 1 rankingWidely usedWidely used	0.86B (UNet)	7.0 GB approx. ▣ Edge · ≤10 GB ✓ Fits on your computer disk 4.3 GB 1× 8–12 GB (approx.)	CreativeML OpenRAIL-M
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven Massive use · 1.6 M HF downloads/mo · 1 leaderboard (AA Text-to-Image) · from 2022-10 · Adoption: Widely used (⬇ 1.6M/mo HF · 1 ranking) (leaderboards: AA Text-to-Image; released: 2022-10-01; data from 2026-10-09) Arena AA Text-to-Image: #168 · Elo 456 (±13, 95% CI) · 10,388 votes Score: AA text-to-image arena: #168 of 168. Memory: disk 4.3 GB (FP32) · Min. VRAM 7.0 GB approx. · fits on: 1× 8–12 GB (approx.) Checkpoint pruned EMA 4.3 GB UNet (diffusers) 3.4 GB Disk = pruned EMA checkpoint. No official VRAM figure: estimate = pruned EMA checkpoint × 1.1 + 2 GB (encoders offloaded). License: CreativeML OpenRAIL-M Source: artificialanalysis.ai/image/leaderboard/text-to-image · 2026-10-09
Qwen-Image-2.1-Turbo Alibaba	n/d	€0.014$0.016	OpenOpen	Barely testedBarely t. Just publishedNo usage dataNo data	7.1B (transformer)	18.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 32.4 GB 1× 24 GB (approx.)	Qwen Research (non-commercial)
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. Just released · no leaderboard of its own yet · Adoption: No usage data (Just published) (released: 2026-10-09; data from 2026-10-09) Score: 8 steps (same 7B). Turbo API $0.016/img (Singapore). Weights on HF; AA Elo n/a. Research license (non-commercial). Memory: disk 32.4 GB (BF16) · Min. VRAM 18.0 GB approx. · fits on: 1× 24 GB (approx.) Transformer BF16 (7.1B) 14.2 GB Encoder (text) BF16 17.5 GB VAE BF16 0.7 GB Requirement: Model card: no VRAM figure; same 7B architecture as Qwen-Image-2.1. source Repo ~32.4 GB (encoder 17.5 + transformer 14.2 + VAE 0.7). Research/evaluation-only license. License: Qwen Research (non-commercial) Source: docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen-image-2 · 2026-10-09
Iris-3B Sperid Labs	n/d	self-host	OpenOpen	Barely testedBarely t. ⬇ 18/mo HF · 0 rankingsLittle usedLittle used	3B (+ encoder Qwen3-VL-4B)	15.0 GB approx. ▣ 1 consumer GPU ✓ Fits on your computer disk 11.9 GB 1× 24 GB (approx.)	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: Barely testedBarely t. No independent benchmark · 18 HF downloads/mo · 4 days · Adoption: Little used (⬇ 18/mo HF · 0 rankings) (released: 2026-10-05; data from 2026-10-09) Score: New (Oct 8). 3B without VAE; matches Qwen-Image on OneIG (according to its authors). Not in the arena yet. Memory: disk 11.9 GB (F32) · Min. VRAM 15.0 GB approx. · fits on: 1× 24 GB (approx.) Text to image F32 11.9 GB Requirement: Model card: NVIDIA GPU with CUDA; ~12 GB of weights for text to image; no VRAM figure. source The depth/ and upscaler/ folders add 12 GB more each; the Qwen3-VL-4B encoder is downloaded separately. License: Apache 2.0 Source: huggingface.co/speridlabs/iris-3b · 2026-10-09

10 open-source models and variants in video · 5 fit on your computer

Quality (Elo) vs price per second (open only)

Up = better · left = cheaper · the dashed line is the efficient frontier.

5 models · 2 on the frontier · hover over or tap a point to see its name and values

ClosedOpen weightsEfficient frontier

Swipe to see the whole chart

Price with the cheapest API that offers it; on your computer the cost is your hardware and electricity.

Table: Video (open only)

10 models and variants · click ▸ on a row to see sources, memory and notes

Sort

Model	EloAA (audio) ⇅	€/sUSD/s1080p ⇅	WeightsWeights⇅	MaturityMad.⇅	Paramstotal / act.	Memory ⇅VRAM · disco	License
MiniMax H3 (768p) MiniMax	1137 #4	€0.071/s$0.080/s	OpenOpen	ProvenProven ⬇ 3.6M/mo HF · 4 rankingsWidely usedWidely used	~33B (transformer, BF16 on HF)	25.0 GB approx. ▣ 1 consumer GPU disk 66.3 GB 1× 32 GB (approx.)	MiniMax H3 Community License
Featured: ★ Best open · on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 3.6 M HF downloads/mo · 4 leaderboards · 71 days · Adoption: Widely used (⬇ 3.6M/mo HF · 4 rankings) (leaderboards: AA T2V v2.0, LMArena T2V, LMArena I2V, AA I2V v1.0; released: 2026-07-30; data from 2026-10-09) Arena AA T2V v2.0: with audio #4 (Elo 1137) · no audio #3 (Elo 1116) Score: Best open weights and very cheap. Memory: disk 66.3 GB (BF16 (transformer)) · Min. VRAM 25.0 GB approx. · fits on: 1× 32 GB (approx.) Transformer BF16 66.3 GB Transformer podado BF16 40.2 GB Transformer podado FP8 21.0 GB Transformer podado W6A8 16.0 GB Transformer podado GGUF Q4_K 11.4 GB Encoder Qwen3-VL 32B BF16 51.5 GB NVFP4 encoder (does not require Blackwell) 15.7 GB Requirement: Model card: inference example with --num-gpus 4; no official VRAM figure. source Official repo (diffusers): several pipelines (FL2VA, Ref2VA) with a 32B encoder; the pruned/quantized variants are from Comfy-Org and unsloth. No official VRAM figure; the official example uses --num-gpus 4. License: MiniMax H3 Community License Source: artificialanalysis.ai/video/leaderboard/text-to-video?audio-output=tru · 2026-10-08
LTX-2.5 Fast Lightricks	946 #23	€0.12/s$0.13/s	OpenOpen	ProvenProven ⬇ 1.7M/mo HF · 2 rankingsWidely usedWidely used	22B	32.0 GB official ▣ 1× 32–48 GB disk 42.0 GB 1× 32 GB (official min.) · 80 GB recommended	LTX-2.x Community
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.7 M HF downloads/mo · 2 leaderboards (AA T2V v2.0, AA I2V v1.0) · 59 days · Adoption: Widely used (⬇ 1.7M/mo HF · 2 rankings) (leaderboards: AA T2V v2.0, AA I2V v1.0; released: 2026-08-11; data from 2026-10-09) Arena AA T2V v2.0: with audio #23 (Elo 946) · no audio #22 (Elo 937) Score: Same weights repo as LTX-2.5 Pro (AA links to Lightricks/LTX-2.5); in the API it's the fast mode. Memory: disk 42.0 GB (BF16 (transformer)) · Min. VRAM 32.0 GB official · fits on: 1× 32 GB (official min.) · 80 GB recommended Transformer BF16 42.0 GB Transformer int8 (ComfyUI) 21.5 GB Transformer NVFP4 (Blackwell) 18.7 GB GGUF Q4_K_M (community) 15.7 GB Requirement: LTX docs: NVIDIA GPU with 32 GB+ of VRAM, 32 GB of RAM; recommended A100 80 GB or H100. source Plus Gemma 4 12B encoder (26.3 GB BF16 · 15.4 GB int8). Low VRAM: --quantization fp8-cast --offload cpu (model card). License: LTX-2.x Community Source: artificialanalysis.ai/video/leaderboard/text-to-video?audio-output=tru · 2026-10-09
LTX-2.5 Pro Lightricks	943 #24	€0.15/s$0.17/s	OpenOpen	ProvenProven ⬇ 1.7M/mo HF · 2 rankingsWidely usedWidely used	22B	32.0 GB official ▣ 1× 32–48 GB disk 42.0 GB 1× 32 GB (official min.) · 80 GB recommended	LTX-2.x Community
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 1.7 M HF downloads/mo · 2 leaderboards (AA T2V v2.0, AA I2V v1.0) · 59 days · Adoption: Widely used (⬇ 1.7M/mo HF · 2 rankings) (leaderboards: AA T2V v2.0, AA I2V v1.0; released: 2026-08-11; data from 2026-10-09) Arena AA T2V v2.0: with audio #24 (Elo 943) · no audio #23 (Elo 933) Score: Free license under $10M in revenue; gated repo. Memory: disk 42.0 GB (BF16 (transformer)) · Min. VRAM 32.0 GB official · fits on: 1× 32 GB (official min.) · 80 GB recommended Transformer BF16 42.0 GB Transformer int8 (ComfyUI) 21.5 GB Transformer NVFP4 (Blackwell) 18.7 GB GGUF Q4_K_M (community) 15.7 GB Requirement: LTX docs: NVIDIA GPU with 32 GB+ of VRAM, 32 GB of RAM; recommended A100 80 GB or H100. source Plus Gemma 4 12B encoder (26.3 GB BF16 · 15.4 GB int8). Low VRAM: --quantization fp8-cast --offload cpu (model card). License: LTX-2.x Community Source: artificialanalysis.ai/video/leaderboard/text-to-video?audio-output=tru · 2026-10-08
LTX-2.3 Pro Lightricks	892 #27	€0.071/s$0.080/s	OpenOpen	ProvenProven ⬇ 971k/mo HF · 2 rankingsWidely usedWidely used	22B	16.0 GB community ▣ 1 consumer GPU ✓ Fits on your computer disk 46.1 GB 1× 16 GB tight (FP8 + offload) · 24 GB comfortable	LTX-2 Community
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 971k HF downloads/mo · 2 leaderboards (AA T2V v2.0, AA I2V v1.0) · from 2026-03 · Adoption: Widely used (⬇ 971k/mo HF · 2 rankings) (leaderboards: AA T2V v2.0, AA I2V v1.0; released: 2026-03-05; data from 2026-10-09) Arena AA T2V v2.0: with audio #27 (Elo 892) · no audio #29 (Elo 856) Score: Fits in 16–24 GB with FP8/offload. Memory: disk 46.1 GB (BF16 (transformer)) · Min. VRAM 16.0 GB community · fits on: 1× 16 GB tight (FP8 + offload) · 24 GB comfortable Transformer dev BF16 46.1 GB FP8 transformer (official) 29.1 GB NVFP4 transformer (official, Blackwell) 21.7 GB GGUF Q4_K_M distilled (community) 14.3 GB Requirement: Community guide: minimum 16 GB with FP8 distilled + sequential offload; 24 GB BF16 distilled with offload; 48 GB+ BF16 without offload. source Plus Gemma 3 12B encoder (24.4 GB BF16 · 13.2 GB FP8 community). 48 GB+ for BF16 without offload (community guide). License: LTX-2 Community Source: artificialanalysis.ai/video/leaderboard/text-to-video?audio-output=tru · 2026-10-08
LTX-2.3 Fast Lightricks	855 #28	€0.054/s$0.060/s	OpenOpen	ProvenProven ⬇ 971k/mo HF · 2 rankingsWidely usedWidely used	22B	16.0 GB community ▣ 1 consumer GPU ✓ Fits on your computer disk 46.1 GB 1× 16 GB tight (FP8 + offload) · 24 GB comfortable	LTX-2 Community
Featured: on the efficient frontier (nothing gives more performance for less money) Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 971k HF downloads/mo · 2 leaderboards (AA T2V v2.0, AA I2V v1.0) · from 2026-03 · Adoption: Widely used (⬇ 971k/mo HF · 2 rankings) (leaderboards: AA T2V v2.0, AA I2V v1.0; released: 2026-03-05; data from 2026-10-09) Arena AA T2V v2.0: with audio #28 (Elo 855) · no audio #28 (Elo 857) Score: Same weights repo as LTX-2.3 Pro (AA links to Lightricks/LTX-2.3); in the API it's the fast mode. Memory: disk 46.1 GB (BF16 (transformer)) · Min. VRAM 16.0 GB community · fits on: 1× 16 GB tight (FP8 + offload) · 24 GB comfortable Transformer dev BF16 46.1 GB FP8 transformer (official) 29.1 GB NVFP4 transformer (official, Blackwell) 21.7 GB GGUF Q4_K_M distilled (community) 14.3 GB Requirement: Community guide: minimum 16 GB with FP8 distilled + sequential offload; 24 GB BF16 distilled with offload; 48 GB+ BF16 without offload. source Plus Gemma 3 12B encoder (24.4 GB BF16 · 13.2 GB FP8 community). 48 GB+ for BF16 without offload (community guide). License: LTX-2 Community Source: artificialanalysis.ai/video/leaderboard/text-to-video?audio-output=tru · 2026-10-09
HunyuanVideo 1.5 Tencent	LMArena T2V 1169 (n.º 38/48)	self-host	OpenOpen	ProvenProven ⬇ 684k/mo HF · 3 rankingsWidely usedWidely used	8.3B	14.0 GB official (offload) ▣ 1 consumer GPU ✓ Fits on your computer disk 33.3 GB Official min. 14 GB (with offload) · Wan2GP from 6 GB	Tencent Hunyuan Community (excludes EU, UK and South Korea; >100M MAU requires a license)
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 684k HF downloads/mo · 3 leaderboards · from 2025-11 · Adoption: Widely used (⬇ 684k/mo HF · 3 rankings) (leaderboards: LMArena T2V, LMArena I2V, AA I2V v1.0; released: 2025-11-20; data from 2026-10-09) Score: License NOT valid in the EU/UK/South Korea. Memory: disk 33.3 GB (FP32 (transformer, per variant)) · Min. VRAM 14.0 GB official (offload) · fits on: Official min. 14 GB (with offload) · Wan2GP from 6 GB Official FP32 (each 480p/720p, T2V/I2V variant) 33.3 GB Comfy-Org FP16 16.6 GB Comfy-Org FP8 scaled 8.3 GB GGUF Q8_0 (community) 9.0 GB GGUF Q4_K_M (community) 5.1 GB Encoder Qwen2.5-VL 7B FP16 16.6 GB Encoder Qwen2.5-VL 7B FP8 9.4 GB Requirement: Model card: «Minimum GPU Memory: 14 GB (with model offloading enabled)». Wan2GP (community): from 6 GB. source The official repo stores each transformer in FP32 (33.3 GB) and has 11 variants (480p/720p, T2V/I2V, distilled, SR 1080p). Distilled 480p I2V step: ~75 s per video on an RTX 4090 (model card). Not in the AA T2V v2.0 arena (with audio). AA I2V v1.0 (no audio): Elo 1134, #66 of 83 (LTX-2.3 Pro: 1158, #61). LMArena: T2V 1169 (#38/48), I2V 1198 (#40/48). No VBench for 1.5 (the original 13B HunyuanVideo: 83.2%). API on fal: $0.075/s at 480p. License: Tencent Hunyuan Community (excludes EU, UK and South Korea; >100M MAU requires a license) Source: lmarena.ai/leaderboard/text-to-video · 2026-10-09
CogVideoX1.5-5B Zhipu (Z.ai)	VBench 82.0 %	self-host	OpenOpen	NewNew ⬇ 2.5k/mo HF · 1 rankingModerately usedModerately used	5.6B	10.0 GB official (offload) ▣ 1 consumer GPU ✓ Fits on your computer disk 11.1 GB Official min. 10 GB (diffusers BF16 optimized) · INT8 from 7 GB	CogVideoX License (commercial use: free registration; >1M visits/month, separate license)
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew Only 1 leaderboard (VBench) · 3k HF downloads/mo · from 2024-11 · Adoption: Moderately used (⬇ 2.5k/mo HF · 1 ranking) (leaderboards: VBench; released: 2024-11-02; data from 2026-10-09) Score: Old (Nov 2024); not in the AA arena or LMArena. Memory: disk 11.1 GB (BF16 (transformer)) · Min. VRAM 10.0 GB official (offload) · fits on: Official min. 10 GB (diffusers BF16 optimized) · INT8 from 7 GB Transformer BF16 (T2V) 11.1 GB Transformer BF16 (I2V) 11.1 GB Encoder T5-XXL 19.1 GB Requirement: Model card: diffusers BF16 «from 10GB» and INT8 (torchao) «from 7GB» with optimizations; SAT BF16 76 GB; multi-GPU BF16 24 GB. 5 s in ~550 s on an H100. source Not in any AA or LMArena arena. VBench T2V: 82.0% (Wan2.2-A14B: 82.6%, LTX-2: 79.4%). 1360×768, 5–10 s at 16 fps, English-only prompts. Successor CogVideoX-3 API only (no weights). License: CogVideoX License (commercial use: free registration; >1M visits/month, separate license) Source: huggingface.co/spaces/Vchitect/VBench_Leaderboard · 2026-10-09
Wan 2.2 A14B Alibaba	LMArena T2V 1132 (n.º 42/48)	self-host	OpenOpen	ProvenProven ⬇ 297k/mo HF · 4 rankingsWidely usedWidely used	27B (2 experts of 14B) / 14B	80.0 GB official ▣ 1× 80 GB disk 114 GB 1× 80 GB (official, with offload) · less with FP8/GGUF in ComfyUI	Apache 2.0
Weights: OpenOpen 🤗 Hugging Face Maturity: ProvenProven 297k HF downloads/mo · 4 leaderboards · from 2025-07 · Adoption: Widely used (⬇ 297k/mo HF · 4 rankings) (leaderboards: LMArena T2V, LMArena I2V, AA I2V v1.0, VBench; released: 2025-07-28; data from 2026-10-09) Score: Last Wan with open weights (Wan 2.5+ API only). Memory: disk 114 GB (FP32 (2 experts × 57.15 GB)) · Min. VRAM 80.0 GB official · fits on: 1× 80 GB (official, with offload) · less with FP8/GGUF in ComfyUI Official FP32 (high + low noise) 114 GB Comfy-Org FP16 (2 × 28.58) 57.2 GB Comfy-Org FP8 scaled (2 × 14.29) 28.6 GB GGUF Q8_0 (2 × 15.40) 30.8 GB GGUF Q4_K_M (2 × 9.65) 19.3 GB Encoder UMT5-XXL FP16 11.4 GB Encoder UMT5-XXL FP8 6.7 GB Requirement: Model card: the 1-GPU command (720p, --offload_model --convert_model_dtype) «can run on a GPU with at least 80GB VRAM». source Only one expert (14B) works at each step; ComfyUI loads one at a time, so with FP8 (14.3 GB per expert) or GGUF Q4 (9.65 GB) VRAM drops a lot (no official figure). Not in AA T2V v2.0. AA I2V v1.0 (no audio): 1107, #69 of 83. LMArena: T2V 1132 (#42), I2V 1170 (#43). VBench T2V 82.6% (84.2% with an optimized prompt). Smaller sibling TI2V-5B (720p, fits in 24 GB). License: Apache 2.0 Source: lmarena.ai/leaderboard/text-to-video · 2026-10-09
Kandinsky 6.0 Video Pro Kandinsky Lab	No AA Elo yet (new)	self-host	OpenOpen	NewNew ⚠ little use · New on HFLittle usedLittle used	29B	79.0 GB approx. ▣ 1× 80 GB disk 90.0 GB Large GPU / multi-GPU or offload (transformer ~70 GB BF16)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew By default: new open · ~270 HF downloads/mo · no AA arena yet · Adoption: Little used (⚠ little use · New on HF) Score: Open MIT · 29B · audio sync; vendor: beats open LTX-2.5 in visual/voice quality. Memory: disk 90.0 GB (BF16) · Min. VRAM 79.0 GB approx. · fits on: Large GPU / multi-GPU or offload (transformer ~70 GB BF16) Pro BF16 (repo) 90.0 GB Requirement: Gearbrain: Pro 29B; block offloading lets you run it on a home PC at the cost of speed. source HF total: transformer 69.73 + text_encoder 16.58 + rest ~3.7 GB. Vendor (side-by-side): better than open LTX 2.5 in visuals and speech; in image animation it beats Veo 3.1 Fast (closed) according to Kandinsky. Only open MiniMax H3 is ahead. No public API price. License: MIT Source: huggingface.co/kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers · 2026-10-10
Kandinsky 6.0 Video Lite Kandinsky Lab	No AA Elo yet (new)	self-host	OpenOpen	NewNew ⚠ little use · New on HFLittle usedLittle used	3B	16.0 GB community ▣ 1 consumer GPU ✓ Fits on your computer disk 27.7 GB 1× 16–32 GB (official presets Lite; offload)	MIT
Weights: OpenOpen 🤗 Hugging Face Maturity: NewNew By default: new open · ~200 HF downloads/mo · no AA arena yet · Adoption: Little used (⚠ little use · New on HF) Score: Open MIT · 3B · 5 s clips with audio sync; fits on a consumer GPU. Memory: disk 27.7 GB (BF16) · Min. VRAM 16.0 GB community · fits on: 1× 16–32 GB (official presets Lite; offload) Lite BF16 (repo) 27.7 GB Lite distill transformer 6.4 GB Requirement: Gearbrain/Kandinsky: Lite adapted to consumer RTX cards with 16, 24 and 32 GB presets; block offloading lets you run it at home. source HF total: transformer 7.48 + text_encoder 16.58 + audio_vae 1.71 + vae 0.99 + clip 0.49 + vocoder 0.49 GB. T2AV + TI2AV, 44 kHz, lip-sync. Text encoder Qwen2.5-VL (~16.6 GB) shared. Presets 16/24/32 GB (Gearbrain). Distill 10-step available. License: MIT Source: huggingface.co/kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers · 2026-10-10

Keep exploring

How to install themLM Studio, Ollama… to use open-source models.Open and closedWhat the difference is and when each one makes sense.What VRAM isYour graphics card's memory, explained.