SmellBench - Resolution Rate: leaderboard
SmellBench (2026) scores coding-agent CLIs on repairing 65 expert-validated architectural code smells in scikit-learn. This board is the percentage of all 65 smells that PyExamine no longer detects after the agent run, including smells removed as side effects of other fixes; higher is better.
Metric: Share of all 65 smells (%) that PyExamine no longer detects after the run (a resolution may be a side effect of work on another task), 65 expert-validated hard-severity PyExamine architectural smells in scikit-learn, off-the-shelf coding agent CLIs with the SmellBench agent skill and GEPA-optimized task packets, default reasoning settings, one run per agent; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.3 Codex Spark | 47.7 | |
| 2 | GPT-5.4 | 32.3 | #135 |
| 3 | GPT-5.3 Codex | 30.8 | #124 |
| 4 | Gemini 3.1 Pro (Preview) | 29.2 | #99 |
| 5 | Claude Sonnet 4.6 | 27.7 | #173 |
| 6 | Claude Opus 4.6 | 27.7 | #113 |
| 7 | Claude Haiku 4.5 | 27.7 | #583 |
| 8 | GPT-5.1 Codex Max | 27.7 | #207 |
| 9 | Mistral Vibe + Devstral 2 | 16.9 | |
| 10 | Gemini 3.1 Flash Lite (Preview) | 9.2 | #417 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=smellbench-resolution-rate · How It Works · Data refreshed daily, snapshot 2026-11-21.