Every comparison is built from the same dataset that powers the release timeline: officially published benchmark scores (GPQA Diamond, SWE-Bench Verified, MMMU, and more), model specifications, and release dates. Head-to-head verdicts only count benchmarks that both models actually report, so a model never “wins” by default on a test the other never took.

Looking for a specific model first? Browse the latest releases or jump to a company page to see every model we track from that lab.