Skip to content

Benchmark

mu-MATH F1

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
1.50

Score distribution

74%
Mean
75%
Median
9.8%
Std dev
58.8%
10th percentile
85.8%
90th percentile
48.7%
Min
48.7%93.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.50 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

#ModelScore
51Claude 3.5 SonnetGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.8%
52Qwen2.5 72B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.7%
53GPT 4oGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.6%
54Mistral Large Instruct 2411Qwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.5%
55GPT-4o minimu_math, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.3%
56Qwen2.5 7B InstructLlama-3.1-70B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.3%
57Claude 3.5 SonnetGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.2%
58Gemini 1.5 FlashQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH71.2%
59Claude 3.5 SonnetQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.8%
60Qwen2.5 72B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.5%
61GPT-4o miniGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.4%
62Gemini 1.5 FlashGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.1%
63GPT-4o miniGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.6%
64Llama 3.1 70B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.4%
65GPT-4o miniQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.3%
66Qwen2.5 7B Instructmu_math, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.3%
67Qwen2.5 Math 72B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.3%
68Qwen2.5 7B InstructGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.1%
69Qwen2.5 7B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH68.3%
70Qwen2.5 Math 72B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH68.2%
71Qwen2.5 Math 7B InstructGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH63.8%
72Qwen2.5 Math 7B InstructLlama-3.1-70B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH63.8%
73Qwen2.5 7B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH62.4%
74Qwen2.5 Math 7B Instructmu_math, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH61.9%
75Llama 3.1 70B Instructmu_math, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH61%
76Qwen2.5 Math 7B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH59.7%
77Llama 3.1 70B InstructGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH58.8%
78Qwen2.5 Math 7B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH57.2%
79Llama 3.1 70B InstructLlama-3.1-70B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH57%
80Llama 3.1 70B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH56%
81Llama 3.1 8B InstructGemini-1.5-Pro, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH55.5%
82Llama 3.1 8B Instructmu_math, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH52%
83Llama 3.1 8B InstructGPT-4o, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH51.2%
84Llama 3.1 8B InstructLlama-3.1-70B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH49.2%
85Llama 3.1 8B InstructQwen2.5-72B-Instruct, F1, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH48.7%

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)