Atlas

Benchmarks

← All benchmarks

AA-Robotics v0.1

Agents · 2026-10-14

Artificial Analysis's AA-Robotics v0.1 research preview has a model drive a physical Hiwonder 6-axis robot arm through an HTTP API with four camera views. Eight tabletop tasks run from pressing a button and placing a toy to pouring a drink, tidying blocks, shape matching, writing, tower building and measuring a box. Each model runs inside its vendor's coding agent. Every task gets one zero-shot attempt, scored 0-100 with partial credit for incomplete tasks, and the score is the mean over the eight tasks. Eight single attempts make it a diagnostic.

Top models (higher is better)

ModelScore
GPT-6 Astra69.4
GPT-6.1 Sol64.6
Gemini 4 Argon55.3
Claude Opus 5.548.8
Claude Fable 5.146.3
Loading Atlas data…