AlgoVeri (Dafny): leaderboard
Metric: Full-mark rate (%) on AlgoVeri's 77 classical-algorithm vericoding problems in Dafny: the implementation and proof must verify against the fixed specification within a 15-round compiler-feedback repair budget and pass an LLM semantic filter that rejects specification gaming; single attempt (open-weight rows are the mean of 10 independent attempts); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 40.26 | #78 |
| 2 | Qwen 3 235B A22B 2507 Instruct | 18.31 | #291 |
| 3 | Qwen 3 Next 80B A3B (Thinking) | 17.27 | #306 (Qwen 3 Next 80B A3B) |
| 4 | GPT-OSS-120B | 13.51 | #330 |
| 5 | Devstral 2 | 6.1 | #394 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=algoveri-dafny · How It Works · Data refreshed daily, snapshot 2026-10-11.