MGSM
Math · 2022-10-06
MGSM is a multilingual grade-school mathematics benchmark built from manually translated GSM8K problems in ten languages. Scores are reported as exact-answer accuracy.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.5 | 95.2 |
| Opus 4.1 | 94.4 |
| Sonnet 4.5 | 94.3 |
| GPT-5.2 | 94.0 |
| Gemini 3 Pro Preview | 93.9 |
| Opus 4 | 93.8 |
| O4 Mini | 93.4 |
| Gemini 3 Flash Preview | 93.3 |
| Sonnet 4 | 93.0 |
| Claude 3.7 Sonnet | 93.0 |
| GPT-5.1 | 93.0 |
| GPT-5 | 92.8 |
| Claude 3.5 Sonnet (Oct 2024) | 92.6 |
| GPT-5 Mini | 92.6 |
| Qwen3-235B-A22B | 92.5 |
| Llama 4 Maverick Instruct | 92.4 |
| DeepSeek-R1 | 92.3 |
| Haiku 4.5 | 92.1 |
| DeepSeek-V3 | 92.1 |
| Qwen3 Max Preview | 92.1 |
| gpt-oss-120b | 92.0 |
| Qwen3 Max (rolling alias) | 91.8 |
| o3 | 91.7 |
| DeepSeek-V3-0324 | 91.7 |
| Claude 3.5 Sonnet (June 2024) | 91.6 |
| Grok 3 | 91.3 |
| o3-mini | 91.3 |
| Llama 3.3 70B Instruct | 91.1 |
| Kimi K2 Instruct | 90.9 |
| Grok 4 | 90.9 |
| DeepSeek-V3.2 | 90.9 |
| Grok 4 Fast | 90.9 |
| Mistral Medium 3 | 90.9 |
| GLM-4.5 | 90.8 |
| GPT-4o (2024-08-06) | 90.7 |
| Grok 3 Mini | 90.4 |
| GPT-4o (2024-11-20) | 90.4 |
| Kimi K2 Thinking | 90.1 |
| Gemini 2.5 Flash Preview (09-2025) | 89.9 |
| GLM 4.6 | 89.7 |