← All benchmarks

Codeforces - DeepSeek-reported competitor percentile

Code

Percentile of human Codeforces competitors that a model outperforms, as computed by DeepSeek's internal Codeforces evaluation in its DeepSeek-V3 and DeepSeek-R1 reports (problems from recent Division 2 contests graded with expert-crafted test cases). The contest set and rating procedure are DeepSeek-internal and change between reports, so the row is diagnostic.

Top models (higher is better)