Model results

DrivingBench

Choose a shared evaluation protocol to compare model scores. Versions, sources and configurations stay separate.

Atlas snapshot captured 10/19/2026, 10:20:28 AM UTC. This is an import timestamp, not the evaluation date. Coverage may be incomplete.

Official source

← All evaluation groups

DrivingBench

Source: DrivingBench (reviewed publisher results)
Version: 2026-09-v1-cone-course-best-of-3
Measurement date: Not reported
Metric: % course progress · Higher is better
Evaluation scope: Best progress of up to 3 attempts in one continuous chat; 45% along the course centerline within 4 m; attempt 3 reached 45% (attempts 1–2: 9%, 10%), DNF; harness=Claude Code, reasoning=medium; real 2022 Toyota Corolla via comma four and openpilot, three MCP tools (observe, set_motion, stop_now), human on the brake; single physical run per attempt, not a repeated-trial success rate; result date not supplied; captured 2026-09-23.

Fewer than two comparable models. No ranking is assigned. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.

Claude Fable 5.1
anthropic/claude-fable-5-1
45% course progress
View modelEvidence independent