← Leaderboard
π0.7
VLAPhysical Intelligence · Apr 16, 2026 · built on π0.6
Steerable ~5B VLA prompted with language plus episode metadata and subgoal images, trained with knowledge insulation on top of the π0.6 architecture. Results are self-reported and largely qualitative (zero-shot cross-embodiment laundry folding on a bimanual UR5e, espresso making that the paper says matches RL-fine-tuned specialists), not on a public benchmark.
- Type
- VLA
- Parameters
- 5B
- Backbone
- Gemma 3 4B + MEM video-history encoder + 860M action expert
- Weights
- Closed
- License
- –
- Commercial use
- Unclear
- Released
- Apr 16, 2026
- Robots
- bimanual mobile manipulator, static bimanual (BiPi), bimanual UR5e
- SO-101
- –
- LeRobot
- –
- Inputs
- RGB · P · L
- VRAM
- –
Benchmark results
0 of 13 benchmarksNo benchmark results recorded yet.
Not reported on 13 benchmarks
Wrong or missing number? Report an error