China's Free AI Models Are Catching Up Fast
New data shows China's open AI is only 7 months behind US leaders—what that means for you.
Someone built an open-source AI leaderboard called whatmodel.sh, and it's packed with data: 22,113 benchmark results across 987 benchmarks, pulled from 1,011 sources, and every single number links back to where it was published.
The chart itself focuses on text models: 18,755 results covering 1,211 models on 617 benchmarks. Of those, 14,436 are independent evaluations — papers whose authors ran the models themselves, mostly CC BY papers from the ACL Anthology and arXiv, plus Epoch AI, the LMArena dataset, EvalPlus, the EleutherAI harness and others. The remaining 7,677 are scores labs reported for their own models, labelled as lab-reported.
Everything is open: the dataset is CC BY-SA 4.0 and the scoring code is MIT, available at github.com/jakemanger/whatmodel, with CSV downloads at whatmodel.sh/data/manifest.json. Every number links back to its source, and each source's licence is recorded. There's an interactive version too — click any model to see its scores and sources at whatmodel.sh/race.
- China's free AI models are only 7 months behind top US paid models, according to a new leaderboard.
- This could lead to cheaper or free AI tools for everyone, as competition increases.
- But open models may be less safe and require more technical skill to use.
Why It Matters
You might soon get near-top-tier AI for free, saving money and giving you more control.