SelfBench - earendil-works/pi: leaderboard
Metric: Task success (%) on 51 tasks rebuilt from the repository's merged pull requests (hidden tests). Source: selfbench.dev. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 Sol (High) | 84.31 |
| 2 | GPT-6.1 Sol (High) | 82.35 |
| 3 | GPT-6.1 Sol (Low) | 82.35 |
| 4 | Claude Opus 5.5 (High) | 82.35 |
| 5 | GPT-6 Sol (Low) | 78.43 |
| 6 | Claude Opus 5.5 (Low) | 76.47 |
| 7 | Kimi K3 (High) | 74.51 |
| 8 | GPT-6 Luna (High) | 68.63 |
| 9 | GPT-6 Luna (Low) | 56.86 |
Interactive version: theaggregate.ai/benchmark?slug=selfbench-earendil-works-pi · How It Works · Data refreshed daily, snapshot 2026-10-03.