AA-Robotics v0.1 leaderboard

Artificial Analysis research preview in which frontier language models, run as coding agents (Codex, Claude Code, Antigravity), drive a real arm through an HTTP API and four cameras. Only general LLMs are evaluated, not VLAs; a human grader scores each attempt against the task’s pass criteria.

Real robot · Hiwonder NexArm (6-axis)Report an error
79Best · GPT-6.2 Astra
9Models ranked
OpenSaturation
Nov 18, 2026Numbers checked
#ModelTypeScoreDateSource
1GPT-6.2 AstraOpenAILLM controller79Nov 17, 2026
2GPT-6.2 SolOpenAILLM controller78Nov 17, 2026
3Claude Fable 5.2AnthropicLLM controller75Nov 17, 2026
4GPT-6 AstraOpenAILLM controller69.4Oct 14, 2026
5GPT-6.1 SolOpenAILLM controller64.6Oct 14, 2026
6Gemini 4 HydrogenGoogle DeepMindLLM controller58, self-reported by the authors, not a leaderboard entryNov 17, 2026
7Gemini 4 ArgonGoogle DeepMindLLM controller55.3Oct 14, 2026
8Claude Opus 5.5AnthropicLLM controller48.8Oct 14, 2026
9Claude Fable 5.1AnthropicLLM controller46.3Oct 14, 2026

Metric: Mean human-graded score (0–100, partial credit) over 8 zero-shot tabletop tasks, one attempt each. One result per model: RoboFYI re-run, then official leaderboard, then paper, then third party. Runs that deviate from the official protocol are listed on the model page but not ranked.

Wrong or missing number? Report an error