Benchmark
GSM8K
Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.
- Models and systems measured
- 104
- Scores recorded
- 104
- Sources
- 5
- IQ weight
- 0.30
Score distribution
- 74%
- Mean
- 87.6%
- Median
- 28%
- Std dev
- 25.2%
- 10th percentile
- 95.1%
- 90th percentile
- 0%
- Min
This benchmark has a weight of 0.30 for chat models, measured over 97 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | Score |
|---|---|---|
| 1 | MiMo V2.5 Pro· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 99.6% |
| 2 | Kimi K2 Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 97.3% |
| 3 | o1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 97.1% |
| 4 | GPT-4.5 Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 97% |
| 5 | Llama 3.1 405B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.8% |
| 6 | Llama 3.1 405B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.8% |
| 7 | Claude 3.5 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.4% |
| 8 | Gemma 3 27B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.9% |
| 9 | Qwen2.5 32B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.9% |
| 10 | Qwen2.5 72B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.8% |
| 11 | DeepSeek V2.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.1% |
| 12 | Llama 3.1 70B Instruct8-shot · chain of thought· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95.1% |
| 13 | Claude 3 Opus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95% |
| 14 | Nova Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.8% |
| 15 | Qwen 2.5 14B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.8% |
| 16 | DeepSeek Coder V2 Instruct· Epoch AI Benchmarking Hub data | 94.5% |
| 17 | Nova Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.5% |
| 18 | Gemma 3 12B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.4% |
| 19 | Qwen 3 235B A22B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.4% |
| 20 | qwen2.5-coder-14b-instruct· Epoch AI Benchmarking Hub data | 94.2% |
| 21 | Llama 3.2 90B Vision Instruct· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 93.1% |
| 22 | Mistral Large 2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93% |
| 23 | Qwen 2.5 Coder 32B Instruct· Epoch AI Benchmarking Hub dataValues from 2 sources
| 93% |
| 24 | DeepSeek V4 Pro· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 92.6% |
| 25 | Granite 4.1 8B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 92.5% |
| 26 | Claude 3 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 92.3% |
| 27 | Nova Micro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 92.3% |
| 28 | Qwen2.5 7B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 91.6% |
| 29 | Llama 3.1 Nemotron 70B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 91.4% |
| 30 | GPT-4o mini· Epoch AI Benchmarking Hub data | 91.3% |
| 31 | Qwen2 72B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 91.1% |
| 32 | Qwen 2.5 Coder 32B· Epoch AI Benchmarking Hub data | 91.1% |
| 33 | Phi-3 Medium 4K Instruct· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 91% |
| 34 | Gemini 1.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 90.8% |
| 35 | GPT 4· Epoch AI Benchmarking Hub data | 90% |
| 36 | Qwen 2 (72B)· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 89.5% |
| 37 | DeepSeek V3· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 89.3% |
| 38 | Gemma 3 4B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 89.2% |
| 39 | Claude 3 Haiku· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 88.9% |
| 40 | phi-3.5-moe-instruct· Epoch AI Benchmarking Hub dataValues from 2 sources
| 88.7% |
| 41 | Qwen2.5-Omni 7B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 88.7% |
| 42 | qwen2p5-coder-14b· Epoch AI Benchmarking Hub data | 88.7% |
| 43 | Phi 4 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 88.6% |
| 44 | deepseek-coder-v2-lite-instruct· Epoch AI Benchmarking Hub data | 87.6% |
| 45 | Jamba 1.5 Large· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87% |
| 46 | Claude Instant 1.2· Epoch AI Benchmarking Hub data | 86.7% |
| 47 | Qwen2.5 Coder 7B Instruct· Epoch AI Benchmarking Hub dataValues from 2 sources
| 86.7% |
| 48 | Gemini 1.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.2% |
| 49 | deepseek-coder-v2-base· Epoch AI Benchmarking Hub data | 85.8% |
| 50 | Phi-3 Mini 4K Instruct· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.7% |
Showing 1–50 of 98 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | Score |
|---|---|
| DeepSeek V4.1 Flash BaseEM · 8-shot· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93% |
| MiMo V2 Flash Baseofficial Base table · 8-shot· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 92.3% |
| Qwen3.5-397B-A17B-BaseQwen3.5 release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93.7% |
| Qwen3.7-Plus-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93% |
| Qwen3.8-27B-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93.2% |
| Qwen3.8-Flash-Next-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93.3% |
Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Data by LLM Stats · Hugging Face official benchmark leaderboards; licence varies by benchmark · OpenEvals/leaderboard-data, MIT (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.