XRepoTest - PHP - Test Pass Rate: leaderboard
XRepoTest (arXiv 2608.25939v1, Appendix Table A.4): repository-level unit-test generation in real open-source PHP projects, standard single-turn setting with greedy decoding, with generated tests executed in containers. Share of all focal functions whose generated test suite executes and passes, percent. Higher is better.
Metric: Passing generated test suites (%; of all focal functions). Source: arxiv.org. Saturation forecast: Around October 2028. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 26.93 | #148 |
| 2 | Claude Haiku 4.5 (20251001) | 14.86 | #264 |
| 3 | GLM-5 | 13.16 | #149 |
| 4 | DeepSeek V4 Pro | 13 | #112 |
| 5 | GPT-5.2 | 10.06 | #115 |
| 6 | MiniMax-M2.7 | 5.57 | #270 |
| 7 | Codestral-22B-v0.1 | 5.57 | #850 |
| 8 | DeepSeek V3.2 | 5.42 | #210 |
| 9 | Qwen3 Coder Next | 5.11 | #331 |
| 10 | Yi-Coder-9B-Chat | 3.56 | #1244 |
| 11 | GPT-OSS-120B | 3.41 | #346 |
| 12 | Kimi K2.5 | 3.25 | #152 |
| 13 | Llama 3.3 70B Instruct | 2.01 | #541 |
| 14 | Qwen 3 8B | 0.62 | #678 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=xrepotest-php-test-pass-rate · How It Works · Data refreshed daily, snapshot 2026-11-05.