Back to BenchmarksFrontier
GSM8K
Grade school math word problems requiring multi-step arithmetic reasoning
Leaderboard
(20 models)| Rank | Model | Score | Stderr |
|---|---|---|---|
| 1 | DeepSeek V3 | 94.50 | — |
| 2 | Qwen2.5-Max | 94.20 | — |
| 3 | Qwen2.5-Coder-32B-Instruct | 93.00 | — |
| 4 | GPT-4.1 | 92.00 | — |
| 5 | gpt-4o-mini-2024-07-18 | 91.30 | — |
| 6 | Phi-3-mini-4k-instruct | 88.70 | — |
| 7 | Claude Opus 4.5 | 86.70 | — |
| 8 | Gemma 3 27B | 84.90 | — |
| 9 | Mistral Large | 84.20 | — |
| 10 | Gemini 1.5 Flash | 82.40 | — |
| 11 | Llama 3.1 405B | 82.40 | — |
| 12 | Mixtral-8x7B-v0.1 | 74.40 | — |
| 13 | Llama-2-70b-hf | 69.60 | — |
| 14 | Qwen 3 235B | 61.30 | — |
| 15 | Llama-2-7b | 58.70 | — |
| 16 | gpt-3.5-turbo-1106 | 57.77 | — |
| 17 | falcon-180B | 54.40 | — |
| 18 | Mistral-7B-v0.1 | 54.40 | — |
| 19 | gemma-7b | 46.40 | — |
| 20 | Yi-6B | 44.90 | — |
Data source: Epoch AI, “Data on AI Benchmarking”. Published at epoch.ai
Licensed under CC-BY 4.0