ClarifyCodeBench: leaderboard
ClarifyCodeBench (Peking University, 2026) deletes one to three details from 419 LiveCodeBench v6 problems and lets the model ask clarification questions, answered by the benchmark when a GPT-4o judge matches them to the annotated key questions, before it writes code. Score is the pass@1 of the final code on the hidden tests, in percent; higher is better.
Metric: Pass@1 (%) on the hidden LiveCodeBench v6 tests of the code written after interactive clarification of an ambiguous requirement (419 tasks with one to three deleted requirement details; a GPT-4o judge matches each question to the annotated key questions and the environment returns the deleted detail; at most 6 rounds, temperature 0). Source: arxiv.org. Saturation forecast: Around September 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 41.1 | #167 |
| 2 | DeepSeek V3.2 (Non-reasoning) | 39.1 | #406 (DeepSeek V3.2) |
| 3 | Claude Sonnet 4.5 | 38.2 | #250 |
| 4 | Claude Sonnet 4.5 (Thinking) | 34.3 | #250 (Claude Sonnet 4.5) |
| 5 | Gemini 2.5 Flash (Non-reasoning) | 29.8 | #453 (Gemini 2.5 Flash) |
| 6 | Qwen 3 235B A22B (Non-reasoning) | 27.7 | #623 (Qwen 3 235B A22B) |
| 7 | GPT-4o | 27.2 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clarifycodebench · How It Works · Data refreshed daily, snapshot 2026-11-13.