CodeSpecBench-Func - Correctness: leaderboard

CodeSpecBench (2026) asks models to write executable precondition and postcondition functions that specify the intended behaviour of a program; the function-level set uses 2,494 LeetCode problems with online-judge-validated valid and invalid tests. Correctness, the share of specifications that accept every valid test, in percent; higher is better.

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: Estimated already saturated. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 2.5 Pro88.1#276
2Claude Sonnet 4.5Thinking85#261
3Gemini 2.5 Flash83.7#463
4QwQ-32B83.1#907
5GPT-OSS-120B82.6#704
6GPT-5 Mini82#364
7Qwen 3 32B81.9#972
8Qwen 3 14B80.1#1269
9Qwen 3 8B75.5#1873
10GPT-575.1#178
11GPT-OSS-20B73.7#1183
12DeepSeek V3.2Non-reasoning72.9#419
13Qwen 3 4B71.5#2348
14Qwen 3 1.7B14.7#5860
15Qwen 3 0.6BNon-reasoning8.4#8350

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=codespecbench-func-correctness · How It Works · Data refreshed daily, snapshot 2026-11-21.