← Leaderboard
GR-RL
VLAByteDance Seed · Dec 1, 2025 · built on GR-3
GR-3 specialized for long-horizon, high-precision dexterous manipulation with data filtering, augmentation and offline-plus-online real-world reinforcement learning. Results are self-reported: 83.3% success on autonomous shoe lacing vs 45.7% for the GR-3 imitation baseline.
- Type
- VLA
- Parameters
- 5B
- Backbone
- Qwen2.5-VL-3B + DiT (GR-3) with learned critic
- Weights
- Closed
- License
- –
- Commercial use
- Unclear
- Released
- Dec 1, 2025
- Robots
- ByteMini-v2
- SO-101
- –
- LeRobot
- –
- Inputs
- –
- VRAM
- –
Benchmark results
0 of 13 benchmarksNo benchmark results recorded yet.
Not reported on 13 benchmarks
Wrong or missing number? Report an error