TestExplora-Lite (Black Box): leaderboard
Metric: Fail-to-pass rate (%): share of pull requests for which at least one generated test fails before the fix and passes after it; TestExplora-Lite (517 samples from 330 pull requests with high-quality human docstrings); Black Box setting, where the model sees only the documentation and signatures of the functions under test; single model call with DocAgent-written documentation as the oracle; mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 7.67 | #176 |
| 2 | O3 Mini | 5.58 | #266 |
| 3 | O4 Mini | 3.74 | #172 |
| 4 | GPT-4o | 3.61 | #333 |
| 5 | Gemini 2.5 Pro | 2.6 | #145 |
| 6 | Qwen 3 Coder 30B A3B Instruct | 2.3 | #399 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=testexplora-lite-black-box · How It Works · Data refreshed daily, snapshot 2026-10-11.