Benchmark
mu-MATH TPR
Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.
- Models measured
- 17
- Scores recorded
- 85
- Sources
- 1
- IQ weight
- 0.53
Score distribution
- 74.2%
- Mean
- 76.1%
- Median
- 12%
- Std dev
- 57.9%
- 10th percentile
- 90.5%
- 90th percentile
- 45.2%
- Min
This benchmark has a weight of 0.53 for chat models, measured over 17 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 51 | Qwen2.5 Math 7B InstructGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 75% |
| 52 | DeepSeek R1GPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 72.2% |
| 53 | GPT 4oGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 72.1% |
| 54 | DeepSeek R1Qwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 71.1% |
| 55 | GPT 4oLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 71.1% |
| 56 | Mistral Large Instruct 2411Qwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 71% |
| 57 | Qwen2.5 Math 7B InstructQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 71% |
| 58 | Claude 3.5 SonnetGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 70.9% |
| 59 | GPT 4oGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 70.4% |
| 60 | GPT 4omu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 70.1% |
| 61 | Gemini 1.5 FlashGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 67.7% |
| 62 | GPT 4oQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 67.1% |
| 63 | Llama 3.1 70B InstructGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 67.1% |
| 64 | Gemini 1.5 FlashLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 67% |
| 65 | Llama 3.1 70B InstructLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 63.9% |
| 66 | Gemini 1.5 Flashmu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 63.3% |
| 67 | GPT-4o miniGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 63% |
| 68 | Claude 3.5 Sonnetmu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 62.5% |
| 69 | Llama 3.1 70B Instructmu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 62.5% |
| 70 | Llama 3.1 70B InstructGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 61.4% |
| 71 | Gemini 1.5 FlashQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 60.6% |
| 72 | Claude 3.5 SonnetLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 59.8% |
| 73 | GPT-4o miniLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 59.8% |
| 74 | GPT-4o minimu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 59% |
| 75 | Llama 3.1 70B InstructQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 58.7% |
| 76 | Claude 3.5 SonnetQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 58.1% |
| 77 | Claude 3.5 SonnetGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 57.9% |
| 78 | Gemini 1.5 FlashGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 57.9% |
| 79 | GPT-4o miniGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 56.4% |
| 80 | GPT-4o miniQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 56.1% |
| 81 | Llama 3.1 8B InstructGemini-1.5-Pro, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 55% |
| 82 | Llama 3.1 8B Instructmu_math, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 48.7% |
| 83 | Llama 3.1 8B InstructGPT-4o, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 46.4% |
| 84 | Llama 3.1 8B InstructLlama-3.1-70B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 45.4% |
| 85 | Llama 3.1 8B InstructQwen2.5-72B-Instruct, TPR, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH | 45.2% |
Showing 51–85 of 85 results.
Data credits: Toloka, U-MATH and mu-MATH public leaderboard. MIT Space; generated solution outputs retain underlying model terms and are not ingested. (licence)