All AI evaluation results ever conducted are now on Hugging Face model pages

In February 2026, the EEE (Every Eval Ever) project launched under the EvalEval coalition, the first cross-organizational initiative to improve reporting of AI evaluations. Simultaneously, Hugging Face launched Community Evals for decentralized display of benchmarks on its platform. Previously, evaluation results were scattered: published in papers, leaderboards, blogs, and logs, often in different formats, and the same model on the same benchmark could show varying scores (e.g., LLaMA 65B obtained 63.7 and 48.8 on MMLU) due to undocumented settings. EEE introduces a unified JSON schema for recording evaluation results, including data on who conducted the evaluation, which model, access method, generation settings, and metric description. The EEE repository on Hugging Face already contains about 229,000 results for more than 22,000 models and 2,200 benchmarks, extracted from 31 report formats. Now, a converter automatically transforms EEE records into YAML files required by Hugging Face Community Evals and submits them. Results on model pages receive a verified checkmark from EvalEval if the data was submitted via an official organization account. The converter maps fields, checks existing results, and creates a pull request only after user confirmation. Supported benchmarks include MMLU-Pro, GPQA, HLE, and GSM8K.