# MambaAct speeds up vision‑language‑action robots

> The new MambaAct expert cuts inference time by up to 65% while slightly improving success rates on LIBERO benchmarks.

Oossa · 2026-10-19 · https://oossa.com/en/mambaact-speeds-up-vision-language-action-robots

Researchers introduced MambaAct, a smaller action expert for flow‑matching vision‑language‑action (VLA) models. It keeps the visual‑language encoder of the earlier π₀.₅ model but replaces its transformer‑based action part with a compact Mamba module. The change reduces the action expert size by ten‑fold and cuts denoising latency by 64.8% on the LIBERO simulation benchmark and 58.8% on real‑world tests. At the same time, average success on LIBERO improves by about 1 percentage point. The code and model weights are released publicly.

## The facts

- MambaAct is 10× smaller than the previous action expert.
- Latency drops 64.8% on LIBERO and 58.8% in real‑world deployment.

## Why it matters

Roboticists can run VLA models faster on limited hardware, enabling near‑real‑time manipulation.

## Sources & references

1. [MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models](https://arxiv.org/abs/2610.22951) – arXiv cs.RO, 2026-10-19

Last updated: 2026-10-19