Image: drivingbench.com · DrivingBench: frontier language models driving a real car through a cone course
DrivingBench: frontier LLMs driving a real Toyota Corolla
A real car instrumented so language models can drive it on a test course.
Credits
- HN submitter aditya-ramabadr Source ↗
Earliest dated source:
The full story
DrivingBench is a benchmark where frontier language models, including Claude Fable, GPT-6, Grok and Sol, drive a real Toyota Corolla fitted with a comma device through a cone course in an empty car park at low speed. They issue one command at a time through MCP tools, using the car's cameras and telemetry, with a human supervisor ready to brake. A leaderboard shows each model's runs, traces and videos.
How it was done
Tools
Themes
Loading themes…
Sources
From the maker's own write-up or code. We summarise what they report; we haven't tested it ourselves.
- Hacker News: Show HN: DrivingBench – Frontier LLMs Driving a Real Toyota Corolla
The leaderboard, with traces and videos of each run.
- Hacker News discussion (3 points)HN submitter: aditya-ramabadrPosted
The Show HN post.