Skip to content

Benchmark

MLE-Bench

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
20
Scores recorded
34
Sources
3
IQ weight
1.16

Score distribution

42%
Mean
44%
Median
19.7%
Std dev
7.6%
10th percentile
63.9%
90th percentile
1.6%
Min
1.6%66.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.16 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Sonnet 5Gemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.9%
2Gemini 3 Pro PreviewFamou-Agent 2.0· MLE-bench official data64.4%
3Gemini 3.6 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim63.9%
4Gemini 3.6 FlashGemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.63.9%
5Claude Opus 4.6AIBuildAI· MLE-bench official data63.1%
6Gemini 3 Pro PreviewCAIR MARS+· MLE-bench official data62.7%
7Gemini 3 FlashLoongFlow· MLE-bench official data62.7%
8Gemini 3 Pro PreviewMLEvolve· MLE-bench official data61.3%
9Gemini 3 Pro PreviewPiEvolve (Fractal AI Research)· MLE-bench official data61.3%
10Gemini 2.5 ProFamou-Agent 2.0· MLE-bench official data59.6%
11DeepSeek V3.2 SpecialeML-Master 2.0· MLE-bench official data56.4%
12Gemini 3 Pro PreviewCAIR MARS· MLE-bench official data56%
13Gemini 3 Pro PreviewLeeroo· MLE-bench official data50.7%
14Gemini 3.5 FlashGemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.49.7%
15GPT-5 CodexThesis· MLE-bench official data48.4%
16GPT-5.6 LunaGemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.47.6%
17Gemini 2.5 ProCAIR MLE-STAR-Pro-1.5· MLE-bench official data44%
18Gemini 2.5 ProFamou-Agent· MLE-bench official data43.6%
19Grok 4.5Gemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.43.2%
20Gemini 3.1 Pro PreviewGemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.42.6%
21GPT 5Operand ensemble· MLE-bench official data39.6%
22Gemini 3.5 Flash Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim39.2%
23Gemini 2.5 ProCAIR MLE-STAR-Pro-1.0· MLE-bench official data38.7%
24DeepSeek R1InternAgent· MLE-bench official data36.4%
25GPT 5R&D-Agent· MLE-bench official data35.1%
26OpenAI o3AIRA-dojo· MLE-bench official data31.6%
27DeepSeek R1ML-Master· MLE-bench official data29.3%
28O1 PreviewR&D-Agent· MLE-bench official data22.4%
29O1 PreviewAIDE· MLE-bench official data17.1%
30GPT 4oAIDE· MLE-bench official data8.6%
31Claude 3.5 SonnetAIDE· MLE-bench official data7.6%
32GPT 4oOpenHands· MLE-bench official data4.9%
33Llama 3.1 405B InstructAIDE· MLE-bench official data3.3%
34GPT 4oMLAB· MLE-bench official data1.6%

Data credits: MLE-bench (OpenAI), MIT (licence) · 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. · Data by LLM Stats