How we measure it

The score (0–100)

• General: Score = points from 0 to 100 on a battery of hard, independent tests (Artificial Analysis index).
• Coding: Score = % of real coding tasks it solves in an independent test (Terminal-Bench 4.0).
• Images and video: these tests give an Elo score (blind votes between two results). We turn it into a score with the Elo formula: score = 2 × the probability that people prefer it over #1. #1 gets 100; one that wins 4 out of 10 duels against it gets 80.

The simple lists have one row per model with its best version; the variants (max, high, medium…) are in «For the data nerds», at the end of each page.

The three cards

«The most powerful» is the one with the highest score. «The best value» is the one that gives the most score for what it costs to use (it's on the efficient frontier: no other model is better and cheaper at the same time); we review it by hand at each update. Adoption rule: a «Barely tested» model (no independent evaluation) or a «Little used» one can never be «The best value» if there is a well-adopted alternative (widely used, or moderately used if none is widely used); in that case that one is chosen. The same goes for «The best free one (open source)»: if the open model with the highest score is «Barely tested» or «Little used» and there is another more adopted (widely or moderately used) and evaluated open model, the highest-scoring one of those is chosen. This applies in all categories. «The most powerful» is always the one with the highest score, even if it's barely tested; in that case the card warns about it. «The best free one (open source)» is the best model you can download and use without paying and without limits, among those already measured by at least one independent test.

What we call «free». On this site, «free» only means truly free: open-source models that you download and use on your own machine, without paying and without limits. No paid model appears in the «free» lists or cards, even if its app (ChatGPT, Claude, Gemini…) has free use with limits: that is shown as a note on the model's own card («It also has free use with limits in the app, in addition to the paid plans.»). We only say an app has free use with limits when its official page confirms it; otherwise we put «Price not confirmed».

Prices in euros

App prices (ChatGPT, Claude, Gemini, Midjourney, Cursor…) come from their official pricing pages. If the page publishes the price in euros for Spain or the EU, we show it as is. If it only publishes dollars, we show «≈ X €» rounded to the euro, converted at €1 = $1.1206 (ECB reference rate of October 9, 2026), and the original dollar price on hover. So «≈» always means «approximate». Dollar prices on many pages don't include taxes; VAT is added at checkout. If a model is only available via API, we say «API only» and its technical price is in «For the data nerds».

Quality in Spanish

To know which model writes best in Spanish we use the LMArena's Spanish ranking (LMArena · Text Arena, Spanish category (blind votes on questions written in Spanish); data from October 8, 2026. 175,361 votes and 298 models). The top positions are within the margin of error (±) of each other: there is no clear winner. GPT-6.1 Sol and Claude Sonnet 5.5 don't appear in this category yet. It isn't a spelling or style test: it measures which answer users prefer for questions written in Spanish.

Maturity and adoption

Maturity: «Proven» if at least two independent rankings have measured it and it has real usage (or it has been in use for more than 180 days); «New» if it came out less than 30 days ago or only one ranking has measured it; «Barely tested» if no independent ranking has measured it or almost nobody uses it.

Adoption: «Widely used» with more than 100,000 monthly downloads on Hugging Face, a place in the OpenRouter top 20 or more than 1,000 likes; «Moderately used» with more than 5,000 monthly downloads, presence on OpenRouter or more than 100 likes; «Little used» below that. Without public usage data nothing is shown.

Public sources

• Artificial Analysis: intelligence index, Terminal-Bench 4.0, cost per task, image and video arenas.
• LMArena (text, Spanish, web coding, image and video), VBench, MTEB and Decision Index.
• Hugging Face (downloads, likes, actual file sizes and licenses) and OpenRouter (available models and usage ranking).
• Official pricing pages and documentation for each model and tool (memory requirements, plans).
• European Central Bank: euro/dollar reference exchange rate.

Each row of the technical tables links to the specific source of its data. If a figure is not published, «n/a» is shown: nothing is ever estimated without saying so.

Update

It's reviewed twice a day (morning and afternoon, Madrid time): news, new models, ranking positions, downloads and prices. The date and time of the last update appear on the home page.