OpenVLA-OFT
Open weightsStanford
2 of 6 axes published · values checked against their primary sources on 23 Oct 2026
Success
Does it do the trained task? (% of trials, trained setup)
13%
Independent test: clean dish 29%, sponge 23%, unzip 0%, fold 0%. Its own paper reports 97.1% on LIBERO (simulation).
Source: Benchmarking VLAs on ALOHA, arXiv 2511.11298 · Table 3, macro-average
Robustness
Does it survive the cell drifting? (% under the perturbation set)
6%
Unseen objects at random positions: 13%, 11%, 0%, 0%.
Source: Benchmarking VLAs on ALOHA, arXiv 2511.11298 · Table 5 (mean of 4 tasks)
Instruction
Does rewording the instruction break it? (% under paraphrase)
Owed
Not reported.
Recovery
Does it fix its own mistakes? (% of injected failures recovered)
Owed
Not reported.
Speed
Is it as fast as a trained person? (human cycle time ÷ robot cycle time)
Owed
Not reported.
Interventions
How often does a person step in? (per 1,000 cycles)
Owed
Not reported.
Built on OpenVLA-OFT
Published work that starts from OpenVLA-OFT: fine-tunes, add-ons and models trained from it. Their results are listed here, apart from OpenVLA-OFT's own values above, with the source each one comes from.
QbD-VLA
Inference settingOpen codeColorado School of Mines · Northeastern University · USC Institute for Creative Technologies · University of Central Florida · 28 Oct 2026
Shrinks the model by storing most of its numbers at low precision, choosing which parts to compress by testing the robot’s task success in simulation.
- Success91.8%−4.7 pts vs OpenVLA-OFT in the same paper (96.5%)Simulation · LIBERO, 4 suites, 4-bit weights · 800 trials · Grade BReleased per-suite checkpoints. Biggest drop on LIBERO-Long: 80.5% vs 92.5%.
Source: QbD-VLA paper, arXiv 2610.38661
TRIAGE
Inference settingSeoul National University · CMU · 25 Oct 2026
No retraining: skips or reuses parts of each camera frame to run faster on onboard computers.
- Success95.8%−1.3 pts vs OpenVLA-OFT in the same paper (97.1%)Simulation · LIBERO average · Grade BTrades 1.3 points for speed: 443 vs 653 ms per call on a Jetson AGX Orin.
Source: TRIAGE paper, arXiv 2610.34519
Know of one we missed? Send it with its source.