Stanford

OpenVLA-OFT↗

OpenVLA-OFT is a VLA from Stanford, released February 27, 2025. It has no overall rank on the Robot Intelligence Index, which needs both a simulation and a real-robot result. Its best result is #2 of 11 on LIBERO-Plus. OpenVLA fine-tuned with parallel decoding, action chunking, continuous actions and L1 loss.

Release date: Feb 27, 2025

Developer
Stanford
Type
VLA
Model weights
Open
License
MIT
Parameters
7.5B
Pricing
Self-hosted
Benchmarks
4
Overall rank
Not ranked
Needs a real-robot result to rank overall.
Real robots rank
Not ranked
No real-robot results yet.
Simulation rank
#34of 49
Rating 1324 · range 16–42
Rating
1369± 41.2
10601989
Cost / task · est.
$0.010
$0.0087$17.74
Time / task · est.
11s
9.5 s1 h 22 min

Per task attempt; the tiles show the median across benchmarks. Self-hosted compute is priced at one H100 at $3.29 per hour (Lambda on-demand, Oct 2026).

Other models from Stanford

Inference

LIBERO (1 image): 72.9 ms per 8-action chunk, 109.7 Hz throughput on A100; with wrist image+proprio 112 ms / 71.4 Hz. ALOHA OFT+ (3 images, chunk 25): 321 ms, 77.9 Hz.