CodeSpecBench-Func - Correctness: leaderboard
Metric: Correctness (%), the share of generated executable pre/postcondition specifications that accept every valid test, on CodeSpecBench-Func (2,494 LeetCode problems; 50 valid and 50 invalid inputs plus valid and model-generated buggy outputs per problem, validated by the online judge); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 88.1 |
| 2 | Claude Sonnet 4.5 (Thinking) | 85 |
| 3 | Gemini 2.5 Flash | 83.7 |
| 4 | QwQ-32B | 83.1 |
| 5 | GPT-OSS-120B | 82.6 |
| 6 | GPT-5 Mini | 82 |
| 7 | Qwen 3 32B | 81.9 |
| 8 | Qwen 3 14B | 80.1 |
| 9 | Qwen 3 8B | 75.5 |
| 10 | GPT-5 | 75.1 |
| 11 | GPT-OSS-20B | 73.7 |
| 12 | DeepSeek V3.2 (Non-reasoning) | 72.9 |
| 13 | Qwen 3 4B | 71.5 |
| 14 | Qwen 3 32B (Non-reasoning) | 65.8 |
| 15 | Qwen 3 14B (Non-reasoning) | 63.1 |
Interactive version: theaggregate.ai/benchmark?slug=codespecbench-func-correctness · How It Works · Data refreshed daily, snapshot 2026-10-07.