Live-kBench-2512 - Equivalent Patch Rate: leaderboard
Metric: Equivalent patch rate (%): share of the 534 Linux kernel crashes of Live-kBench-2512 (Syzbot bugs reported since April 2024 and fixed between June 2024 and December 2025); the mini-SWE-agent scaffold patches the kernel with crash-resolution feedback (the kEnv environment compiles and runs the reproducer) under a $5.6 budget per bug; Gemini rows are the mean of three runs at temperature 1, the Claude row one run at temperature 0, with a crash-resolving patch that a Gemini 3 Flash judge (majority of nine votes) rates structurally and logically identical to the developer's fix; higher is better. Source: arxiv.org. Saturation forecast: Around August 2028. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 19.85 | #79 |
| 2 | Gemini 3 Pro | 14.67 | #77 |
| 3 | Gemini 3 Flash | 8.18 | #93 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=live-kbench-2512-equivalent-patch-rate · How It Works · Data refreshed daily, snapshot 2026-10-11.