Skip to content

Benchmark

AIME 2024

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

Models measured
50
Scores recorded
60
Sources
2
IQ weight
0.61

Score distribution

75.7%
Mean
80.4%
Median
18.4%
Std dev
48.1%
10th percentile
93.3%
90th percentile
13.1%
Min
13.1%96.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.61 for chat models, measured over 48 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51DeepSeek R1 Distill Qwen 1.5B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim52.7%
52QwQ 32B Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim50%
53GPT-4.1 mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim49.6%
54GPT 4.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim48.1%
55GPT OSS 20bofficial model card Table 3 · reasoning low · no tools · pass@1· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.42.1%
56O1 Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim42%
57DeepSeek V3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim39.2%
58GPT-4.5 Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim36.7%
59GPT-4.1 nano· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim29.4%
60GPT 4o· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim13.1%

Showing 51–60 of 60 results.

Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.