GSO-Bench: leaderboard
102 software optimization challenges across 10 codebases in 5 languages. Models must improve runtime efficiency to match expert developers. Includes hack detection to prevent deceptive optimizations.
Metric: Opt@1 (%). Source: gso-bench.github.io. Saturation forecast: Around January 2027. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5.1 | 88.24 | #9 |
| 2 | GPT-6 Astra | 79.41 | #6 |
| 3 | Claude Fable 5 | 78.43 | #16 |
| 4 | GPT-5.6 Sol | 76.47 | #19 |
| 5 | Claude Opus 4.8 | 47.06 | #37 |
| 6 | Claude Opus 4.7 | 44.12 | #49 |
| 7 | Claude Opus 4.6 | 41.18 | #63 |
| 8 | GPT-5.5 | 40.2 | #31 |
| 9 | Claude Sonnet 5 | 37.25 | #57 |
| 10 | GPT-5.4 | 31.37 | #79 |
| 11 | GPT-5.2 | 27.45 | #110 |
| 12 | Claude Opus 4.5 | 26.47 | #83 |
| 13 | Gemini 3.1 Pro (Preview) | 22.55 | #58 |
| 14 | Gemini 3 Pro | 18.63 | #80 |
| 15 | Claude Sonnet 4.5 | 14.71 | #142 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=gso-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.