Atlas

Benchmarks

← All benchmarks

EmbodiedLab Fine Manipulation (T0 success)

Agents · 2026-10-26

The seven EmbodiedLab experiments whose primary capability dimension is fine manipulation and contact (four basic, three advanced), such as a simple pendulum, a spring launcher and two acid-base titrations. Same T0 interface, scoring gates and averaging as the full benchmark; the score is the success rate on these experiments, in percent.

Top models (higher is better)

ModelScore
GPT-6.1 Sol50.0
Claude Opus 5.542.9
GPT-6 Astra28.6
Kimi K328.6
DeepSeek V4.1 Flash 101714.3
GLM-5.3 Flash0.0
Loading Atlas data…