← Leaderboard

DeepSeek V4.1 Flash (PyRUA) + Galaxea G0.5

Agent system
DeepSeek model; SO-101 Bench harness · Oct 27, 2026 · built on Galaxea G0.5

The same PyRUA-Lean code-as-policy harness with DeepSeek V4.1 Flash as the planner and Galaxea G0.5 for grasps.

Type
Agent system
Parameters
—
Backbone
–
Weights
API
License
–
Commercial use
Unclear
Released
Oct 27, 2026
Robots
SO-101
SO-101
Yes
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
SO-101 BenchLow-cost arm
#2 of 1166.8%, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks

Speed per task

SettingPer taskPer planner callCallsCostSource
SO-101 Bench sim, 916 episodes · A40, 2 agent workers per GPU–––USD 0.024Oct 27, 2026 ↗

Runs on

Wrong or missing number? Report an error