Skip to content

Benchmark

U-MATH overall accuracy

Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.

Models measured
30
Scores recorded
90
Sources
1
IQ weight
1.61

Score distribution

44.8%
Mean
43.9%
Median
22.3%
Std dev
16%
10th percentile
80.7%
90th percentile
9%
Min
9%93.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.61 for chat models, measured over 30 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Llama 3.2 90B Vision Instructu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH41.8%
52Claude 3.5 Sonnetu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH40.7%
53Llama 3.1 70B Instructu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH39.6%
54Claude 3.5 Sonnetu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH38.7%
55Llama 3.2 90B Vision Instructu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH37.2%
56Phi 4u_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH37%
57o3 Miniu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH34.5%
58Llama 3.1 70B Instructu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH34.3%
59Llama 3.1 8B Instructu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH33.7%
60GPT 4ou_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH33.5%
61DeepSeek R1u_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH33%
62DeepSeek V3u_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH32.5%
63Qwen2-VL (72B) Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH32.5%
64Claude 3.5 Sonnetu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH30%
65QwQ 32B Previewu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH30%
66Llama 3.1 8B Instructu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH29.5%
67LFM-7bu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH28%
68Qwen2 VL 7b Instructu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH27.1%
69Qwen2 VL 7b Instructu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH26.3%
70GPT-4o miniu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH26%
71LFM-7bu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH25.8%
72Llama 3.2 11B Vision Instructu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH22.9%
73Qwen2 VL 7b Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH22.5%
74Llama 3.2 11B Vision Instructu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH20.4%
75Nvidia Llama 3.1 Nemotron 70B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH19.5%
76Qwen2.5 Math 72B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH18%
77Pixtral 12Bu_math, text, gpt-4o-2024-08-06· U-MATH and mu-MATH17.9%
78Pixtral 12Bu_math, all, gpt-4o-2024-08-06· U-MATH and mu-MATH17.5%
79Llama 3.2 90B Vision Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH16.5%
80Qwen2.5 72B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH16.5%
81LFM-7bu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH16%
82Pixtral 12Bu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH16%
83Qwen2.5 32B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH16%
84Llama 3.3 70B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH13.5%
85Mistral Large Instruct 2411u_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH12%
86Qwen2.5 Math 7B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH11.5%
87Llama 3.1 8B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH11%
88Qwen2.5 7B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH11%
89Llama 3.1 70B Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH10.5%
90Llama 3.2 11B Vision Instructu_math, visual, gpt-4o-2024-08-06· U-MATH and mu-MATH9%

Showing 51–90 of 90 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)