A report published October 7, 2026, described three Axiom engineers—Aditya Ramabadran, Simon Mahns and Tobias Gessler—connecting general-purpose AI models to a 2024 Toyota Corolla for a supervised In-N-Out drive-through attempt and a parking-lot course. In the reported course results, GPT-6 Astra completed the route slowly, Claude Fable 5.1 covered 45%, and Grok covered 11%.
How the Corolla was connected
The setup linked a chat interface to a server connected to several cameras mounted on the windshield and the car’s power steering. During the In-N-Out attempt, a safety driver kept one foot above the brake. The engineers said the project was outside their work at Axiom.
Results on the parking course
DrivingBench was a low-speed cone course in a parking lot, designed to test steering and speed control. GPT-6 Astra completed it slowly; Claude Fable 5.1 reached 45% of the course, and Grok reached 11%. The results describe performance on that course, with a safety driver present for the separate drive-through attempt.
