Explore / News
AI news
What has come out, most recent first, with one sentence on what it means for you.
Saturday, October 10
- DevelopersNewDrex v1.5 (Nace.AI): open 9B decision model · Nace.AI / Hugging Face
What does it mean for you? A small, free model that chooses between options (for example, classifying messages). Mainly of interest to people building applications.
Technical detail
Open weights (RAIL-M) + OpenRouter API at $0.04/1M in. DI 0.3.1 Full 50.84 · public 58.08 (best open ≤10B on public). ~18 GB BF16 / fits on an A10G 24 GB.
- DevelopersNewMicrosoft-Decision-1 (decisions API) · Microsoft
What does it mean for you? Microsoft offers a very cheap service for applications to classify and score data. It's for developers.
Technical detail
Closed typed-scoring model (Qwen3.5-9B base) on Foundry and OpenRouter: $0.042/1M input, output free, ~0.30 s. Not in the Decision Index yet.
- VideoNewKandinsky 6.0 Video Pro (29B) and Lite (3B) · Kandinsky Lab
What does it mean for you? A free and open video generator that adds synchronized sound and voices. The light version runs on a computer with a good graphics card.
Technical detail
Open MIT with synchronized audio and lip-sync (5 s clips). Lite designed for 16–32 GB consumer GPUs; Pro ~70 GB for the transformer alone.
Friday, October 9
- ImagesNewQwen-Image-2.1-Turbo (open, 8 pasos) · Hugging Face / Qwen
What does it mean for you? A faster version of one of the best free image generators. It can be downloaded, but commercial use is not allowed.
Technical detail
Accelerated checkpoint of 2.1 (7B): same architecture, 8 denoising steps and a Turbo API at ~$0.016/image (Singapore). Weights under the Qwen Research license (non-commercial).
Thursday, October 8
- DevelopersNewAnthropic Cyber Mission and OSS Scanner · X (@AnthropicAI)
What does it mean for you? Anthropic reviews the code security of open-source projects that sign up, for free. Useful if you maintain one.
Technical detail
Free vulnerability scans with its best models (incl. Claude Mythos) for open-source projects that sign up via PR; reports without human review (>90% accuracy expected).
- ModelsNewGPT-6.1 Sol Ultrafast · X (@OpenAIDevs)
What does it mean for you? A mode of OpenAI's model that's up to 8 times faster, in exchange for paying quite a lot more. Only worth it if speed matters most.
Technical detail
Ultrafast mode for GPT-6.1 Sol in the API, Codex and ChatGPT Work: up to 8× faster than Sol Standard; official price 6× Standard (≈ €11$12 / ≈ €54$60 per 1M input/output tokens).
- ToolsNewVoyager, an open creative tool · X (@anvisha)
What does it mean for you? A free tool that combines several AIs to create videos, graphics and games without coding. It's not a new model.
Technical detail
A «coding assistant» for creatives that combines several models (Opus, DeepSeek…) to make video, graphics and games. It's not a new model.
- ImagesNewMagnific One (formerly Freepik) · PR Newswire
What does it mean for you? Magnific's new image generator keeps your brand's style, but each image uses a lot of credits.
Technical detail
Image model that decides the art direction before generating and applies your Brand Kit. Expensive: 180 credits per 2K image, versus 50 for Flux.2 Pro.
- DevelopersNewJetBrains' Mellum2.1 (open, Apache 2.0) · JetBrains blog
What does it mean for you? A free JetBrains model so an assistant can help you code from your own computer, much better than its previous version.
Technical detail
12B with only 2.5B active, designed for coding agents on your own machine. On SWE-bench Verified it goes from ~2 to ~47 versus Mellum2.
- ToolsNewClaude Motion (Anthropic) · The Verge
What does it mean for you? Turns reports and charts into animated videos you can edit. For now only on team plans (Team and Enterprise).
Technical detail
Turns reports and charts into editable animated explainer videos (MP4); exports to Runway. Beta on Team and Enterprise.
- VideoNewGrok Imagine Video 1.5 Lite in the API · xAI
What does it mean for you? xAI offers developers a cheap version of its video generator: $0.02 per second of video.
Technical detail
xAI opens grok-imagine-video-1.5-lite in the API: ≈ €0.018$0.02/s (flat official rate in docs.x.ai). The full 1.5 costs ≈ €0.071$0.08/s. Text and image → video.
- ImagesNewIris-3B (open, Apache 2.0) · Sperid Labs
What does it mean for you? A small, free image generator that runs on a computer with a graphics card. There are no independent tests of its quality yet.
Technical detail
Image without VAE, 3B, ~12 GB. Free (self-host); no Elo yet.
- DevelopersNewJEV-27B-VL: #1 in the Decision Index with images · AutoTrust AI
What does it mean for you? An open model that decides better than any other when images are involved (for example, reviewing photos). It's for developers.
Technical detail
AutoTrust (Apache 2.0, 27.8B, 55.6 GB): 69.8 on the vision board, versus 67.7 for #2. #1 in Hugging Face trending.
Wednesday, October 7
- ModelsClaude Haiku 5.5 (Anthropic) · Anthropic
What does it mean for you? A much cheaper and smarter version than the previous one. Useful if you use AI a lot and want to spend little.
Technical detail
Index 43 (max) versus 17 for Haiku 4.5, at ≈ €0.089$0.10/≈ €0.45$0.50 per 1M tokens: 90% cheaper. Note: x5 with prompts over 100K tokens.
- DevelopersLiquid AI d1 (open, for devices) · Liquid AI blog
What does it mean for you? Tiny models that classify or moderate texts instantly, even on small devices. For developers.
Technical detail
d1-3B and d1-omni-600M: they don't write word by word, they decide in a single pass (classifying, moderating). 8 ms on an RTX 4090, 50 ms on a Jetson Orin Nano.
- VideoVidu Q4 Preview (video up to 4K with audio) · Unite.AI
What does it mean for you? A video generator that reaches 4K with sound and clips of up to 16 seconds. Paid, with a price similar to Runway's.
Technical detail
Clips up to 16 s. ≈ €0.11$0.12/s at 1080p (same as Runway Gen-4.5), ≈ €0.040$0.045/s at 540p and ≈ €0.35$0.39/s a 4K.
- Developerspplx-embed-v2-late 0.6B/9B (Perplexity, MIT) · Perplexity
What does it mean for you? A free model to search inside documents, images and PDFs. It's for people building search engines; it's not a chat.
Technical detail
Multi-vector embeddings for text, images and PDFs without OCR; index with 9B and query with 0.6B.
Tuesday, October 6
- ImagesNano Banana 2.1 (Google) · Google DeepMind
What does it mean for you? Google's new image generator: almost as good as the best and much cheaper.
Technical detail
#4 in the image arena (Elo 1160) at ≈ €0.030$0.034 per image: almost the quality of GPT Image 2.5 for a sixth of the price.
- ModelsMistral Large 4 Preview (weights on Oct 27) · Mistral
What does it mean for you? Mistral's new European AI improves, but it's still behind the best open options. Its weights will be published on October 27.
Technical detail
AA index 38 and $1.13/task at list price ($0.68/$2.09 in preview, −50%). 1.05T parameters (52B active). Below MiMo-V2.6-Pro and GLM-5.3, which are already open.
- DevelopersOpenAI Decisions API, a Jev rival · Fortune
What does it mean for you? OpenAI launches a service for applications to make decisions (yes/no, pick an option, score) with AI. It's for developers.
Technical detail
Yes/no, choice and score decisions with GPT-6 Luna at $0.10/1M input (Jev: $0.042); accepts images. Not in the Decision Index.
- DevelopersEmbeddingGemma 2 (Google, Apache 2.0) · Google
What does it mean for you? A small, free Google model to search by meaning; it works even on modest machines. For developers.
Technical detail
740M multimodal embeddings, 8k context, 1.5 GB (0.18 GB in Q4); ~0.2–0.6 GB of RAM quantized.
- DevelopersOpenAI Decisions API (GPT-6 Luna) (OpenAI)
What does it mean for you? OpenAI launches a decision model for developers. Developers only; you'll find it in: Only via API.
Monday, October 5
- DevelopersPerplexity releases pplx-decider-v1.1-27b: #1 in the Decision Index, ahead of Jev · Hugging Face (Perplexity)
What does it mean for you? Perplexity releases for free the best model for automated decisions, ahead of the paid options. For developers.
Technical detail
Open decision model (Apache 2.0, Qwen3.8-27B base). In Decision Index 0.3 it scores 62.8 versus 60.1 for Jev; fits on 1 80 GB GPU.
- ModelsLiquid AI d1-3B (Liquid AI)
What does it mean for you? Liquid AI launches a language model. Free (open source); you'll find it in: Download it (open source).
- DevelopersPerplexity Decider v1.1 27B (Perplexity)
What does it mean for you? Perplexity launches a decision model for developers with a score of 63/100 in its category. Free (open source); you'll find it in: Download it (open source).
- DevelopersLiquid AI d1-omni-600M (Liquid AI)
What does it mean for you? Liquid AI launches a decision model for developers with a score of 9/100 in its category. Free (open source); you'll find it in: Download it (open source).
Saturday, October 3
- DevelopersTorchcast Decision 27B (Torchcast AI)
What does it mean for you? Torchcast AI launches a decision model for developers with a score of 60/100 in its category. Free (open source); you'll find it in: Download it (open source).
- Developersezjev 4B s2 (everettjf)
What does it mean for you? everettjf launches a decision model for developers with a score of 47/100 in its category. Free (open source); you'll find it in: Download it (open source).
Friday, October 2
- DevelopersBlink v0.3 26B-A4B (Pixilab.ai)
What does it mean for you? Pixilab.ai launches a decision model for developers with a score of 58/100 in its category. Free (open source); you'll find it in: Download it (open source).
Thursday, October 1
- ModelsLing 3.1 Flash (InclusionAI)
What does it mean for you? InclusionAI launches a language model with a score of 41/100 in its category. Developers only; you'll find it in: Only via API.
Wednesday, September 30
- ModelsGemini 4 Argon (Google)
What does it mean for you? Google launches a language model with a score of 53/100 in its category. Developers only; you'll find it in: Only via API.
- ImagesIdeogram 4.5 (High) (Ideogram)
What does it mean for you? Ideogram launches an image generator with a score of 52/100 in its category. Plus: ≈ €18/mo$20/mo; you'll find it in: Ideogram (web and app).
- VideoUtopai X (Utopai)
What does it mean for you? Utopai launches a video generator with a score of 98/100 in its category. Developers only; you'll find it in: Only via API.
- DevelopersFastino GLiDE (28B, no-thinking) (Fastino)
What does it mean for you? Fastino launches a decision model for developers with a score of 60/100 in its category. Developers only; you'll find it in: Only via API.
- DevelopersBespoke Nimble 9B v3 (Bespoke Labs)
What does it mean for you? Bespoke Labs launches a decision model for developers with a score of 55/100 in its category. Free (open source); you'll find it in: Download it (open source).
- DevelopersJEV-27B-VL (AutoTrust) (AutoTrust AI)
What does it mean for you? AutoTrust AI launches a decision model for developers. Free (open source); you'll find it in: Download it (open source).
Monday, September 28
- ModelsClaude Sonnet 5.5 (Anthropic) · Anthropic
What does it mean for you? Almost as smart as Anthropic's best for half the price. A very good choice for everyday use.
Technical detail
Index 56 (max), almost like Opus 5.5 at half the price per token. Since Oct 7, cache reads cost half.
Thursday, September 24
- DevelopersKev 27B (Jared Palmer)
What does it mean for you? Jared Palmer launches a decision model for developers with a score of 59/100 in its category. Free (open source); you'll find it in: Download it (open source).
Wednesday, September 23
- ImagesRecraft V4.1 Flash (Recraft)
What does it mean for you? Recraft launches an image generator with a score of 40/100 in its category. Developers only; you'll find it in: Only via API.
Tuesday, September 22
- ModelsClaude Opus 5.5 (Anthropic)
What does it mean for you? Anthropic launches a language model with a score of 58/100 in its category. Pro: ≈ €18/mo$20/mo; you'll find it in: Claude (Pro plan).
- ModelsGPT-6 Luna (OpenAI)
What does it mean for you? OpenAI launches a language model with a score of 38/100 in its category. Go: ≈ €7/mo$8/mo; you'll find it in: ChatGPT (web and app).
- ImagesMing-Image-0.1-Design (InclusionAI)
What does it mean for you? InclusionAI launches an image generator with a score of 48/100 in its category. Free (open source); you'll find it in: Download it (open source).
- ImagesHunyuanImage 3.5 (Preview) (Tencent)
What does it mean for you? Tencent launches an image generator with a score of 43/100 in its category. Developers only; you'll find it in: Only via API.
- DevelopersDecider 4B (Mapika)
What does it mean for you? Mapika launches a decision model for developers with a score of 40/100 in its category. Free (open source); you'll find it in: Download it (open source).
Monday, September 21
- ModelsMiMo-V2.6-Pro (Xiaomi, MIT) · Hugging Face
What does it mean for you? The smartest open AI right now, and very cheap to use. A good alternative if you don't want to depend on a single provider.
Technical detail
The smartest open-weights model (index 46) and one of the cheapest to run: ≈ €0.12$0.13 per task versus ≈ €0.19$0.21 of GPT-6.1 Sol (medium).
- ModelsGrok 4.7 (xAI (SpaceXAI))
What does it mean for you? xAI (SpaceXAI) launches a language model with a score of 46/100 in its category. Price not confirmed; you'll find it in: Grok (web and app).
- ModelsMiMo-V2.6-Flash (Xiaomi)
What does it mean for you? Xiaomi launches a language model with a score of 38/100 in its category. Free (open source); you'll find it in: Download it (open source).
- DevelopersRune 26B-A4B v3 (Surogate)
What does it mean for you? Surogate launches a decision model for developers with a score of 57/100 in its category. Free (open source); you'll find it in: Download it (open source).
Friday, September 18
- ModelsStep 5 Preview (StepFun)
What does it mean for you? StepFun launches a language model with a score of 44/100 in its category. Developers only; you'll find it in: Only via API.
- DevelopersLaya (Convai Innovations)
What does it mean for you? Convai Innovations launches a decision model for developers with a score of 4/100 in its category. Free (open source); you'll find it in: Download it (open source).
Thursday, September 17
- DevelopersJev 1.13 (System One) (TypeSafe AI)
What does it mean for you? TypeSafe AI launches a decision model for developers with a score of 60/100 in its category. Developers only; you'll find it in: Only via API.
Tuesday, September 15
- VideoBoreal (Creatify)
What does it mean for you? Creatify launches a video generator with a score of 28/100 in its category. Developers only; you'll find it in: Only via API.
Saturday, September 12
- VideoMAGI-2 Preview (0912) (Sand.ai)
What does it mean for you? Sand.ai launches a video generator with a score of 49/100 in its category. Developers only; you'll find it in: Only via API.
Thursday, September 10
- ModelsDeepSeek V4.1 Flash (MIT) · Hugging Face
What does it mean for you? A very fast open model that understands very long texts. A good free option if you need speed.
Technical detail
552B with 16B active and 1M context. Index 39 and the fastest of the big open models (~216 tok/s).
Tuesday, September 8
- ImagesGPT Image 2.5 (OpenAI) · OpenAI
What does it mean for you? The image generator with the best quality today, but also the most expensive per image.
Technical detail
Leader of the image arena (Elo 1198, Sunburst variant) but the most expensive: ≈ €0.19$0.21 per image.
Nothing matches that filter.