Skip to content

Benchmark

mu-MATH NPV

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
0.98

Score distribution

72%
Mean
73.3%
Median
13.7%
Std dev
54.8%
10th percentile
89.3%
90th percentile
37.5%
Min
37.5%95.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.98 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
1o1Llama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH95.8%
2QwQ 32B PreviewLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH95.2%
3O1 MiniLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH93.8%
4DeepSeek R1Llama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH93.6%
5o1Gemini-1.5-Pro, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH91.7%
6Gemini 2.0 Flash ExperimentalLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct, thinking· U-MATH and mu-MATH91.3%
7o1mu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH90.4%
8QwQ 32B Previewmu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH89.7%
9QwQ 32B PreviewQwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH89.3%
10QwQ 32B PreviewGemini-1.5-Pro, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH88.9%
11Qwen2.5 Math 72B InstructLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH88.1%
12o1GPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH87.9%
13Gemini 2.0 Flash Experimentalmu_math, NPV, Qwen/Qwen2.5-72B-Instruct, thinking· U-MATH and mu-MATH87.1%
14Gemini 2.0 Flash ExperimentalGemini-1.5-Pro, NPV, Qwen/Qwen2.5-72B-Instruct, thinking· U-MATH and mu-MATH87%
15Gemini 1.5 ProLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH86.8%
16Qwen2.5 Math 7B InstructLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH86.5%
17Gemini 2.0 Flash ExperimentalGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct, thinking· U-MATH and mu-MATH86.4%
18Qwen2.5 72B InstructLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH85.9%
19Mistral Large Instruct 2411Llama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH85.7%
20Qwen2.5 7B InstructLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH85.3%
21GPT 4oLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH84.9%
22O1 Minimu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH83.8%
23Gemini 1.5 FlashLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH83.3%
24o1Qwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH83.3%
25QwQ 32B PreviewGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH82.1%
26O1 MiniGemini-1.5-Pro, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH81.5%
27Claude 3.5 SonnetLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH80.6%
28GPT-4o miniLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH80.1%
29Gemini 2.0 Flash ExperimentalQwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct, thinking· U-MATH and mu-MATH80%
30DeepSeek R1mu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH79.1%
31O1 MiniGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH77.8%
32O1 MiniQwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH77.4%
33Gemini 1.5 Promu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH76.4%
34Gemini 1.5 ProGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH76.1%
35Qwen2.5 Math 72B Instructmu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH75.2%
36Mistral Large Instruct 2411GPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH74.6%
37DeepSeek R1GPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH74.4%
38Qwen2.5 72B InstructGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.8%
39Qwen2.5 72B Instructmu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.7%
40Qwen2.5 7B InstructGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.7%
41GPT 4oGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.6%
42Mistral Large Instruct 2411mu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.5%
43DeepSeek R1Gemini-1.5-Pro, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH73.3%
44Llama 3.1 70B InstructLlama-3.1-70B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH72.9%
45GPT 4omu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH71.3%
46Qwen2.5 Math 72B InstructGPT-4o, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH71.3%
47Gemini 1.5 ProQwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH71.2%
48Qwen2.5 7B Instructmu_math, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.8%
49DeepSeek R1Qwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH70.3%
50Qwen2.5 Math 72B InstructQwen2.5-72B-Instruct, NPV, Qwen/Qwen2.5-72B-Instruct· U-MATH and mu-MATH69.5%

Showing 1–50 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)