PinchBench — leaderboard
Open-source Kilo.ai benchmark where coding agents complete practical automated tasks such as creating files, researching data, and producing working app artifacts.
Metric: Success Rate (%). Source: pinchbench.com. Status: saturated. 58 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 100 |
| 2 | Mercury 2 | 100 |
| 3 | Qwen 3.7 Max | 93.44 |
| 4 | Grok Build 0.1 | 92.07 |
| 5 | MiMo-V2.5 | 91.87 |
| 6 | Claude Opus 4.8 | 91.76 |
| 7 | Claude Opus 4.7 | 91.58 |
| 8 | DeepSeek V4 Flash | 91.46 |
| 9 | Grok 4.5 | 91.17 |
| 10 | GPT-5.6 Luna | 90.79 |
| 11 | Nemotron 3 Ultra | 90.58 |
| 12 | Claude Haiku 4.5 | 90.4 |
| 13 | Seed 2.0 Lite | 89.71 |
| 14 | MiMo-V2.5-Pro | 89.52 |
| 15 | GPT-5.5 | 88.96 |
Interactive version: theaggregate.ai/benchmark?slug=pinchbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.