Atlas

Benchmarks

← All benchmarks

Codeforces Div. 1 2025 - DeepSeek-V4 expected rating

Code · 2026-04-26

Expected Codeforces rating from DeepSeek's internal benchmark in the DeepSeek-V4 report: 14 Division 1 contests (114 problems, May to November 2025), 32 candidate solutions per problem, 10 sampled in random order as the submission sequence, OpenAI-style penalty scoring against expert test suites, and the per-contest expected rating averaged over contests. The procedure is DeepSeek-internal, so the row is diagnostic.

Top models (higher is better)

ModelScore
DeepSeek-V4-Pro3206
GPT-5.43168
Gemini 3.1 Pro Preview3052
DeepSeek-V4-Flash3052
Loading Atlas data…