Gemini Model Cards - MLE-Bench Partial-30: leaderboard
Google-computed MLE-Bench on the canonical Partial 30 subset of ML engineering tasks, reporting the Average Position Score across two independent runs per problem. Gemini 3.6 Flash and 3.5 Flash-Lite model cards (July 2026) and their methodology PDFs; percent, higher is better; other providers' models at maximum available reasoning settings.
Metric: Average Position Score (%). Source: deepmind.google. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5 | 66.9 | #65 |
| 2 | Gemini 3.6 Flash | 63.9 | #57 |
| 3 | Gemini 3.5 Flash | 49.7 | #66 |
| 4 | GPT-5.6 Luna | 47.6 | #79 |
| 5 | Grok 4.5 | 43.2 | #54 |
| 6 | Gemini 3.1 Pro (Preview) | 42.6 | #64 |
| 7 | Gemini 3.5 Flash Lite | 39.2 | #174 |
| 8 | Gemini 3.1 Flash Lite | 22 | #200 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=gemini-model-cards-mle-bench-partial-30 · How It Works · Data refreshed daily, snapshot 2026-11-05.