SWE-Mutation - Test Repair - Claude Code: leaderboard
Detection of realistic mutation defects by repaired or newly generated tests. Paper v1 Tables 2-3 preserve both tasks and Mini-Swe-Agent/Claude Code scaffolds. RDR is Relative Detection Rate, not patch success; the default mutant generator is Claude Sonnet 4. This series reports Relative Detection Rate (%). Source: https://arxiv.org/html/2605.22175v1.
Metric: Relative Detection Rate (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 81.15 | #148 |
| 2 | Kimi K2 0905 | 74.19 | #292 |
| 3 | GLM-4.6 | 73.54 | #265 |
| 4 | Qwen 3 Coder 480B A35B Instruct | 70.21 | #314 |
| 5 | DeepSeek V3.1 | 68.36 | #275 |
| 6 | Claude 3.7 Sonnet | 66.59 | #259 |
| 7 | GPT-OSS-120B | 39.28 | #346 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swe-mutation-test-repair-claude-code · How It Works · Data refreshed daily, snapshot 2026-11-05.