DrivingBench: leaderboard
Frontier LLMs drive a real 2022 Toyota Corolla through a 127 m parking-lot cone course, steering and braking through a comma four and openpilot from their own agent harness at medium effort. Up to three attempts in one chat; progress is the share of the course centerline reached while staying within 4 m of it.
Metric: Course progress (%, best of up to 3 attempts in one chat). Source: drivingbench.com. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Medium) | 100 | #6 (GPT-6 Astra) |
| 2 | Claude Fable 5.1 (Medium) | 44.88 | #14 (Claude Fable 5.1) |
| 3 | Grok 4.6 (Medium) | 11.02 | #44 (Grok 4.6) |
| 4 | GPT-5.6 Sol (Medium) | 6.3 | #29 (GPT-5.6 Sol) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drivingbench · How It Works · Data refreshed daily, snapshot 2026-11-12.