Asuka-Bench (Claude Code): leaderboard
Asuka-Bench evaluates code agents on 50 web-development tasks given as deliberately underspecified queries, with hidden clarified requirements decomposed into a dependency graph of 784 tasks. This board reports Cumulative weighted task pass rate (%) after three feedback rounds under the Claude Code agent framework, on 50 web-app tasks from underspecified queries verified by a browser UI agent; higher is better.
Metric: Cumulative weighted task pass rate (%) after three feedback rounds under the Claude Code agent framework, on 50 web-app tasks from underspecified queries verified by a browser UI agent; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 89.4 | #161 |
| 2 | GPT-5.4 (Medium) | 88.1 | #123 (GPT-5.4) |
| 3 | GLM-5 | 76.6 | #260 |
| 4 | MiniMax-M2.7 | 76.4 | #518 |
| 5 | Qwen 3.5 Plus | 70.1 | #232 |
| 6 | Seed 2.0 Pro | 51.8 | #211 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=asuka-bench-claude-code · How It Works · Data refreshed daily, snapshot 2026-11-13.