← Leaderboard

InternVLA-A1.5

VLA
Shanghai AI Lab · Jul 6, 2026 · built on InternVLA-A1

Successor to InternVLA-A1 that uses a frozen Wan2.2 video generator only during training, reported on LIBERO, LIBERO-Plus and SimplerEnv WidowX.

Type
VLA
Parameters
2.7B
Backbone
Qwen3.5-2B
Weights
Open
License
CC-BY-NC-SA-4.0
Commercial use
Not allowed
Released
Jul 6, 2026
Robots
WidowX (Bridge), Franka Panda (sim), Aloha-AgileX (sim)
SO-101
–
LeRobot
–

Benchmark results

5 of 13 benchmarks
#28 of 407.1%
#9 of 2485.8%, self-reported by the authors, not a leaderboard entry
RoboTwin 2.0Near saturationSimulation
  • 93.2%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Jul 2026· self-reported, 50 clean + 500 randomized demos· n = 10000· non-standard
93.2%, self-reported by the authors, not a leaderboard entrynot ranked
LIBEROSaturatedSimulation
#10 of 4698.9%, self-reported by the authors, not a leaderboard entry
SimplerEnv WidowXSaturatedSimulation
#6 of 1980.8%, self-reported by the authors, not a leaderboard entry
Not reported on 8 benchmarks
Robustness drop: 98.9% on LIBERO → 85.8% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.

Adoption

updated Nov 17, 2026
GitHub stars
576
GitHub forks
45

Trained on

Wrong or missing number? Report an error