Skip to content

Benchmark

Vals AIME

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

Models measured
81
Scores recorded
86
Sources
1
IQ weight
3.09

Score distribution

65.6%
Mean
82.7%
Median
31.3%
Std dev
15.2%
10th percentile
95.6%
90th percentile
0.417%
Min
0.417%98.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 3.09 for chat models, measured over 81 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Gemini 3.1 Pro Preview· Vals AI98.1%
2GPT 5.2· Vals AI96.9%
3Muse Spark· Vals AI96.9%
4Gemini 3 Pro Preview· Vals AI96.7%
5GPT 5.4· Vals AI96.7%
6Grok 4.20· Vals AI96.5%
7Claude Opus 4.7· Vals AI96.3%
8Claude Opus 4.6· Vals AI95.6%
9Gemini 3 Flash· Vals AI95.6%
10GPT-5.4 mini· Vals AI95.6%
11Kimi K2.5thinking· Vals AI95.6%
12Claude Opus 4.5· Vals AI95.4%
13Qwen3.6 Plus· Vals AI94.6%
14GPT 5· Vals AI93.4%
15GLM 4.7· Vals AI93.3%
16GPT 5.1· Vals AI93.3%
17GLM 4.6· Vals AI92.7%
18GPT OSS 120B· Vals AI92.6%
19Qwen3.5 Flash· Vals AI92.5%
20Claude Sonnet 4.6· Vals AI92.3%
21GLM 5.1· Vals AI91.9%
22Grok 4.1 Fast· Vals AI91.9%
23GLM 5thinking· Vals AI91.7%
24GPT-5 mini· Vals AI91.5%
25Grok 4 Fast· Vals AI91.3%
26MiniMax M2.7· Vals AI91%
27Grok 4· Vals AI90.6%
28GPT 5.4 Nano· Vals AI88.8%
29MiniMax M2.5· Vals AI88.8%
30Claude Sonnet 4.5· Vals AI88.2%
31GLM 4.5· Vals AI86.7%
32o3 Mini· Vals AI86.5%
33GPT OSS 20b· Vals AI86%
34Qwen3.5 Plusthinking· Vals AI86%
35Gemini 2.5 Pro· Vals AI85.8%
36Kimi K2 Thinking· Vals AI85.4%
37OpenAI o3· Vals AI85.3%
38Grok 3 Mini Fast Betahigh, reasoning· Vals AI85%
39DeepSeek V3.2· Vals AI84.6%
40Qwen 3 235B A22B· Vals AI84%
41o4 Mini· Vals AI83.7%
42Magistral Medium· Vals AI83.5%
43Gemini 3.1 Flash Lite· Vals AI83.3%
44Claude Haiku 4.5thinking· Vals AI82.7%
45GPT-5 nano· Vals AI81.2%
46Qwen3 Max· Vals AI81%
47Magistral Small 2509· Vals AI80.7%
48Claude Opus 4.1· Vals AI78.2%
49MiniMax M2.1· Vals AI77.9%
50Claude Sonnet 4thinking· Vals AI76.3%

Showing 1–50 of 86 results.