Atlas

Benchmarks

← All benchmarks

LiveCodeBench Pro

Code · 2025-06-13

LiveCodeBench Pro rates models on competitive-programming problems drawn from Codeforces, ICPC, and IOI contests, captured before editorials or accepted solutions are published. Ratings are Codeforces-scale Elo, recomputed against the original contest standings with a model's penalty time set to the human median. It is a separate benchmark from LiveCodeBench, which reports pass@1 accuracy.

Top models (higher is better)

ModelScore
Gemini 3 Deep Think3298
Gemini 3.1 Pro Preview2887
Gemini 3 Pro Preview2439
GPT-5.2 (2025-12-11)2393
GPT-5.22393
Gemini 3 Flash Preview2316
GPT-5.12243
GPT-5 (2025-08-07)2176
Gemini 2.5 Pro Preview 03-251694
Qwen3 235B A22B Thinking 25071673
Qwen3-Next-80B-A3B-Thinking1603
Sonnet 4.51418
Gemini 2.5 Flash Preview 04-171334
gpt-oss-120b1299
DeepSeek-R1-05281284
Qwen3-30B-A3B1206
DeepSeek-R11161
DeepSeek-V3-03241124
gpt-oss-20b1030
Qwen3-235B-A22B1012
QwQ-32B981
Grok 4 Fast962
GLM-4.5941
GPT-4.1 Mini875
GPT-4.5864
DeepSeek-V3672
GPT-4.1606
Llama 4 Maverick Instruct528
Claude 3.7 Sonnet504
Gemma 3 27B IT402
DeepSeek-R1-Distill-Llama-70B293
GPT-4o (2024-11-20)210
Loading Atlas data…