LiveFMBench - 2025: leaderboard

The contamination-aware split of LiveFMBench (2026): 360 C programs newly collected from SV-COMP 2025, for which a model must write ACSL specifications that let Frama-C prove the embedded assertions. The score is pass@1 (%) over 32 samples under direct prompting, counting a sample only when the proof succeeds and the model left the original code and assertions unchanged; higher is better.

Metric: Pass@1 (%) after the faithfulness check on LiveFMBench-2025 (360 C programs from SV-COMP 2025), direct prompting (instructions plus the C program), 32 samples at temperature 0.7 and top-p 0.95, pass@1 estimated from the 32 samples with the unbiased estimator; a sample counts only when Frama-C/WP (Alt-Ergo and Z3) proves the program and its code and assertions are unchanged (faithfulness check); higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek R1 052815.73#428
2DeepSeek V3.1 TerminusThinking13.64#437
3GLM-4.612.3#498
4Qwen 3 32B7.44#972
5Qwen 3 235B A22B 2507 (Thinking)7.38#466
6Kimi K2 09056.32#570
7DeepSeek V3.2Thinking4.65#419
8Qwen 3 Coder 480B A35B Instruct4.55#652
9Qwen 3 235B A22B 2507 Instruct1.89#616
10DeepSeek V3 (0324)1.28#715
11Llama 3.3 70B Instruct0.25#1385

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

Interactive version: theaggregate.ai/benchmark?slug=livefmbench-2025 · How It Works · Data refreshed daily, snapshot 2026-11-21.