Nvidia's TacitVLA Cuts Vision-Language-Action Inference Latency by 3.4x

1 reports archived1 independent sourcesupdated 11/8/2026, 12:00:00

Synthesis & Latest Updates

1 Sources Cross-Validated

A research team from Nvidia, NYU, UC Berkeley and others introduces TacitVLA, showing that reasoning text in vision-language-action models mainly matters during training rather than at inference. The method matches explicit-reasoning success rates on four manipulation and autonomous-driving benchmarks while cutting latency by up to 3.4x, and sets new records on long-horizon memory tasks.

LATEST/A research team from Nvidia, NYU, UC Berkeley and others introduces TacitVLA, showing that reasoning text in vision-language-action models mainly matters during training rather than at inference. The method matches explicit-reasoning success rates on four manipulation and autonomous-driving benchmarks while cutting latency by up to 3.4x, and sets new records on long-horizon memory tasks.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

机器之心 (微信公众号)T2Nov 8, 12:00·78 pts
Nvidia's TacitVLA Cuts Vision-Language-Action Inference Latency by 3.4x
- Reasoning text accounts for up to 70% of a single VLA model call's latency
- Action decoding need not read the current-step reasoning; learning to write it is sufficient
- Matches explicit reasoning on four benchmarks with up to 3.4x lower latency and up to 17 points higher success at equal speed
View Related Reports & Analysis Original Source