← Leaderboard

Magma

VLA
Microsoft Research · Feb 18, 2025

Multimodal agentic foundation model for both UI navigation and robot manipulation, trained with Set-of-Mark and Trace-of-Mark supervision. It is evaluated zero-shot on SimplerEnv and with few-shot fine-tuning on LIBERO and a real WidowX.

Type
VLA
Parameters
8.9B
Backbone
LLaMA-3-8B + ConvNeXt-XXLarge
Weights
Open
License
MIT
Commercial use
Allowed
Released
Feb 18, 2025
Robots
WidowX 250, Google Robot (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
SimplerEnv WidowXSaturatedSimulation
#25 of 3035.4%, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error