Atlas

Benchmarks

← All benchmarks

EmbodiedLab Long-Horizon (T0 success)

Agents · 2026-10-26

The seven EmbodiedLab experiments whose primary capability dimension is long-horizon multi-stage work (one basic, six advanced), such as a three-terminal circuit black box, a rectifier supply and a two-lens relay. 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
Claude Opus 5.542.9
GPT-6 Astra21.4
GPT-6.1 Sol14.3
DeepSeek V4.1 Flash 10170.0
GLM-5.3 Flash0.0
Kimi K30.0
Loading Atlas data…