OpenCompass Math - Arithmetic (CompassBench 2409): leaderboard

Metric: Score (%). Source: rank.opencompass.org.cn. 30 models tracked.

Top models

#ModelScore
1Qwen 2.5 72B Instruct88.3
2Claude 3.5 Sonnet (20240620)86
3GPT-4o (2024-08-06)85.3
4GPT-4o (2024-05-13)84.7
5DeepSeek V2.582.7
6Qwen 2.5 7B Instruct79.7
7Mistral Large 2 (Jul)79.3
8Gemini 1.5 Pro78.3
9Llama 3.1 405B Instruct FP876.7
10GPT-4o Mini (2024-07-18)76.3
11GLM-4 Plus76
12Llama 3.1 70B Instruct74
13Mistral-Small-Instruct-240972.3
14Step 2 16K72.3
15Gemma 2 27B (IT)71

Interactive version: theaggregate.ai/benchmark?slug=opencompass-math-arithmetic-compassbench-2409 · How It Works · Data refreshed daily, snapshot 2026-09-19.