CodeSpecBench-Repo - Completeness: leaderboard
Metric: Completeness (%), the share of generated executable pre/postcondition specifications that reject every invalid 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 | GPT-5 | 58.6 |
| 2 | Claude Sonnet 4.5 (Thinking) | 57.2 |
| 3 | Gemini 2.5 Pro | 53 |
| 4 | GPT-5 Mini | 45.8 |
| 5 | GPT-OSS-120B | 36.2 |
| 6 | GPT-OSS-20B | 31.6 |
| 7 | DeepSeek V3.2 (Non-reasoning) | 30.2 |
| 8 | QwQ-32B | 25 |
| 9 | Qwen 3 14B | 17.8 |
| 10 | Gemini 2.5 Flash | 17.6 |
| 11 | Qwen 3 14B (Non-reasoning) | 17.4 |
| 12 | Qwen 3 32B | 15.6 |
| 13 | Qwen 3 8B (Non-reasoning) | 10 |
| 14 | Qwen 3 32B (Non-reasoning) | 8.4 |
| 15 | Qwen 3 8B | 7.2 |
Interactive version: theaggregate.ai/benchmark?slug=codespecbench-repo-completeness · How It Works · Data refreshed daily, snapshot 2026-10-07.