CodeSpecBench-Repo - Correctness: leaderboard
Metric: Correctness (%), the share of generated executable pre/postcondition specifications that accept every valid test, on CodeSpecBench-Repo (500 SWE-bench Verified issues; specifications injected around the issue-relevant functions, trigger tests on the fixed and buggy versions with UTBoost-augmented tests); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.5 (Thinking) | 37.4 |
| 2 | Gemini 2.5 Pro | 30.8 |
| 3 | DeepSeek V3.2 (Non-reasoning) | 20.2 |
| 4 | GPT-5 Mini | 19.6 |
| 5 | GPT-5 | 18.8 |
| 6 | QwQ-32B | 10.6 |
| 7 | GPT-OSS-20B | 9.4 |
| 8 | Gemini 2.5 Flash | 9.2 |
| 9 | Qwen 3 32B | 8.8 |
| 10 | GPT-OSS-120B | 8.6 |
| 11 | Qwen 3 14B | 6 |
| 12 | Qwen 3 14B (Non-reasoning) | 4.8 |
| 13 | Qwen 3 32B (Non-reasoning) | 3.2 |
| 14 | Qwen 3 8B | 2 |
| 15 | Qwen 3 8B (Non-reasoning) | 1.6 |
Interactive version: theaggregate.ai/benchmark?slug=codespecbench-repo-correctness · How It Works · Data refreshed daily, snapshot 2026-10-07.