# Kyutai releases REKI, a 2.4 billion-parameter VLA model for robot tasks

> Kyutai, together with Inria and École des Ponts, unveiled REKI, a vision‑language‑action model that adds cheap short‑term memory for robot manipulation.

Oossa · 2026-10-17 · https://oossa.com/en/kyutai-releases-reki-a-2-4-billion-parameter-vla-model-for-robot-tasks

Kyutai, Inria and École des Ponts have launched REKI, a 2.4 billion‑parameter vision‑language‑action (VLA) model. REKI is trained only on publicly available data and includes a lightweight memory of its recent actions. On three out‑of‑distribution benchmarks, only models roughly twice its size performed better. All model weights are publicly released.

## The facts

- Model size: 2.4 billion parameters
- Benchmarks: only larger (~5 billion) models beat it

## Why it matters

Robotics developers can now use an openly available, memory‑enabled model without needing massive proprietary datasets.

## Sources & references

1. [Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation](https://www.marktechpost.com/2026/10/17/meet-kyutai-reki-the-2-4b-vla-model-bringing-efficient-memory-to-robot-manipulation/) – MarkTechPost, 2026-10-17

Last updated: 2026-10-17