CAR-bench — leaderboard
Context-aware reasoning benchmark for agents, focused on following task context, constraints, and environment feedback during multi-step work.
Metric: Avg Pass^3 (%). Source: car-bench.github.io. Status: saturation imminent. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 58 |
| 2 | GPT-5 | 54 |
| 3 | GPT-5.2 | 53 |
| 4 | Claude Opus 4.5 | 52 |
| 5 | Claude Sonnet 4 | 47 |
| 6 | Gemini 2.5 Flash | 41 |
| 7 | Gemini 2.5 Pro | 38 |
| 8 | GPT-4.1 | 37 |
| 9 | Qwen 3 32B | 31 |
Interactive version: theaggregate.ai/benchmark?slug=car-bench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.