CRAB-Bench (Generic User): leaderboard
Metric: Pass rate (%): share of tasks passing both concrete-state and abstract-state verification, with the generic tau2-bench-style user simulator without persona or behaviors; 200 trip-booking tasks (departure and return flight, hotel, attraction) generated from constraint graphs with node and edge distractors, tau-bench framework, pass@1, default inference and reasoning settings, Claude Sonnet 4.6 simulates the user; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3.2 | 75 |
| 2 | GLM-5 | 64 |
| 3 | Claude Sonnet 4.6 | 52 |
| 4 | Qwen3 Coder Next | 45 |
Interactive version: theaggregate.ai/benchmark?slug=crab-bench-generic-user · How It Works · Data refreshed daily, snapshot 2026-09-29.