← All benchmarks

Codeforces - OpenAI-reported Elo

Code · 2024-09-12

Codeforces Elo rating that OpenAI estimates for its own release posts and model cards by having a model compete in simulated Codeforces contests. The o1-era posts allowed up to 10 submissions per problem; later releases used an updated, harder problem set (o3-pro) or a held-out split (gpt-oss), and some runs give the model a terminal tool. Contest set, submission rules and tool access are recorded in score notes; because they change between releases the values are not comparable across releases, so this is diagnostic. Higher is better.

Top models (higher is better)