← Leaderboard
GAM
PolicyKAIST AI · Jun 15, 2026
Geometric Action Model from KAIST AI and ETH Zurich: a policy built on a 3D geometry foundation model rather than a VLM, pretrained on 784K single-arm trajectories.
- Type
- Policy
- Parameters
- 1.4B
- Backbone
- Depth Anything 3 (Giant)
- Weights
- Open
- License
- MIT
- Commercial use
- Allowed
- Released
- Jun 15, 2026
- Robots
- Franka Panda (sim)
- SO-101
- –
- LeRobot
- –
Benchmark results
2 of 13 benchmarksLIBERO-PlusSimulation
- Paper / model card ↗Jun 2026
#8 of 2486.3%, self-reported by the authors, not a leaderboard entry
- Paper / model card ↗Jun 2026
#27 of 4697.6%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Robustness drop: 97.6% on LIBERO → 86.3% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.
Adoption
updated Nov 17, 2026Wrong or missing number? Report an error