allbenched allbenched Allbenched allbenched

Benchmarks / Math

MathArena Final-Answer Competitions

Average over recent final-answer math competitions (AIME, HMMT, BRUMO, SMT, CMIMC and others) run right after each contest. Updated only when a new contest is added, so the newest models can be missing.

CategoryMath
MetricMean accuracy (%)
SourceMathArena (evaluator)
LicenseNone stated
Benchmark pagematharena.ai
StatusScored
Weight0.8 within Math
FitDiscrimination 1.23, noise 0.32

Results

#ModelProviderResultDate
1 GPT-5.5 Extra High OpenAI 94.3% 2026-11-11
2 Claude Opus 4.8 Max Anthropic 91.8% 2026-11-11
3 Kimi K3 Thinking Moonshot AI 87.8% 2026-11-11
4 Gemini 3.1 Pro Preview Google 86.5% 2026-11-11
5 GPT-5.4 Extra High OpenAI 83.1% 2026-11-11
6 Claude Opus 4.6 High Anthropic 78.5% 2026-11-11
7 DeepSeek V4 Pro Max DeepSeek 76.6% 2026-11-11
8 DeepSeek V4 Flash Max DeepSeek 76.6% 2026-11-11
9 Gemini 3.5 Flash Google 76.3% 2026-11-11
10 Claude Opus 4.7 Extra High Anthropic 73.6% 2026-11-11
11 Kimi K2.6 Thinking Moonshot AI 72.9% 2026-11-11
12 GPT-5.2 High OpenAI 72.0% 2026-11-11
13 Gemini 3.6 Flash Google 70.8% 2026-11-11
14 Step 3.7 Flash StepFun 68.5% 2026-11-11
15 GLM-5.2 Zhipu AI 67.6% 2026-11-11
16 Gemini 3 Flash Preview Google 67.6% 2026-11-11
17 GLM-5.1 Zhipu AI 67.1% 2026-11-11
18 Gemini 3 Pro Preview Google 67.0% 2026-11-11
19 Step 3.5 Flash StepFun 66.8% 2026-11-11
20 GLM-5 Zhipu AI 65.7% 2026-11-11
21 Kimi K2.5 Thinking Moonshot AI 62.3% 2026-11-11
22 Grok 4.1 Fast Thinking xAI 60.9% 2026-11-11
23 Nemotron 3 Super 120B A12B NVIDIA 60.4% 2026-11-11
24 DeepSeek V3.2 Thinking DeepSeek 57.7% 2026-11-11
25 Qwen3.5 27B Alibaba 56.7% 2026-11-11
26 Qwen3.5 35B-A3B Alibaba 56.0% 2026-11-11
27 Qwen3.5 9B Alibaba 48.5% 2026-11-11