GPT-6 Astra has gained the ability to drive a car

GPT-6 Astra, run through Codex at medium reasoning effort, completed 100% of the DrivingBench cone course in 5:22, the only model to finish among four frontier models given control of a Toyota Corolla's steering, accelerator, and brakes. Claude Fable 5.1 (Claude Code, medium) reached 45% progress, Grok 4.6 (Cursor, medium) reached 11%, and GPT-5.6 Sol (Codex, medium) reached 6%, with each model allowed up to 3 attempts in one continuous chat. DrivingBench scores attempts by how far along the course centerline a model got while staying within 4 m of it, with a collision preserving only the progress reached before impact.

DrivingBench Equal contribution Can frontier models drive a real car ? We give them control of a Toyota Corolla’s steering, accelerator, and brakes, then evaluate them on a fixed cone course. Leaderboard Up to 3 attempts in one continuous chat. Expand a model to inspect its runs, then click a run to view its trace and video. ModelBest progress Finish time Finish time Attempt 1Attempt 2Attempt 3 GPT-6 AstraCodex · medium100%in 5:22 49% 100% 5:22 — ExpandCollapse GPT-6 Astra Codex · medium 100% 5:22 49% 100% 5:22 — attemptprogressHow far along the course centerline the attempt got while staying within 4 m of it, as a share of the centerline length to the finish zone; a collision keeps the progress reached before it.distanceGPS speed integrated from the first accepted set motion to the end of the attempt's last engagement.finish timeFirst accepted set motion to the end of the attempt's last engagement. Shown for completed runs only; DNF means did not finish.commandsAccepted set motion and stop now calls during the attempt.tokens · costTotal tokens and cost at list prices for the attempt, including the reflection after it. Claude Fable 5.1Claude Code · medium45% 9% 10% 45% ExpandCollapse Claude Fable 5.1 Claude Code · medium 45% 9% 10% 45% attemptprogressHow far along the course centerline the attempt got while staying within 4 m of it, as a share of the centerline length to the finish zone; a collision keeps the progress reached before it.distanceGPS speed integrated from the first accepted set motion to the end of the attempt's last engagement.finish timeFirst accepted set motion to the end of the attempt's last engagement. Shown for completed runs only; DNF means did not finish.commandsAccepted set motion and stop now calls during the attempt.tokens · costTotal tokens and cost at list prices for the attempt, including the reflection after it. Grok 4.6Cursor · medium11% 8% 11% 10% ExpandCollapse Grok 4.6 Cursor · medium 11% 8% 11% 10% attemptprogressHow far along the course centerline the attempt got while staying within 4 m of it, as a share of the centerline length to the finish zone; a collision keeps the progress reached before it.distanceGPS speed integrated from the first accepted set motion to the end of the attempt's last engagement.finish timeFirst accepted set motion to the end of the attempt's last engagement. Shown for completed runs only; DNF means did not finish.commandsAccepted set motion and stop now calls during the attempt.tokens · costTotal tokens and cost at list prices for the attempt, including the reflection after it. GPT-5.6 SolCodex · medium6% 6% 6% 6% ExpandCollapse GPT-5.6 Sol Codex · medium 6% 6% 6% 6% attemptprogressHow far along the course centerline the attempt got while staying within 4 m of it, as a share of the centerline length to the finish zone; a collision keeps the progress reached before it.distanceGPS speed integrated from the first accepted set motion to the end of the attempt's last engagement.finish timeFirst accepted set motion to the end of the attempt's last engagement. Shown for completed runs only; DNF means did not finish.commandsAccepted set motion and stop now calls during the attempt.tokens · costTotal tokens and cost at list prices for the attempt, including the reflection after it. Explore the track View the trajectory replays↓