# Kyutai Releases REKI, a 2.4B Robot-Action Model

Published 2026-10-18 · AI.info Team · AI.info

> Kyutai released REKI, a 2.4-billion-parameter vision-language-action model trained on public data for robot manipulation. Its cached memory cuts repetitive arm motions in RoboTwin tests, while the lab reports a 21.8-millisecond action-chunk inference time on an Nvidia H100.

REKI predicts a robot-action chunk in 21.8 milliseconds on an [Nvidia](https://ai.info/companies/nvidia) H100, according to [Kyutai](https://kyutai.org/reki/)’s compiled inference test. The 2.4-billion-parameter model, released by the Paris-based lab on October 16 with research partners Inria and École nationale des ponts et chaussées, combines camera images and a written instruction to direct a robot. Its model is smaller than the systems it compares against on two benchmarks, Kyutai says, while matching or exceeding their results in several tests.

## Memory built into the action expert

REKI pairs a Qwen3.5-2B vision-language model with a 200-million-parameter flow-matching action expert. The first component encodes camera views and the task instruction; the expert uses that context, along with robot state, to predict movement. Rather than add a separate memory module, REKI uses learned register tokens in the action expert to retain summaries of recent observations.

The model is trained on three-frame windows, with frames taken one second apart. At inference, it caches those summaries and carries them into later action chunks. A three-frame memory adds about 0.1 milliseconds to reading the cached registers and roughly 2 MiB of GPU memory per past frame, Kyutai says.

> “This memory improves success in simulation and on real robots, and makes oscillations between poses about four times rarer.”
>
> Juliette Marrie, Paul Pacaud, Adrien Ramanana Rahary, Shizhe Chen and Moritz Böhle, authors of [“REKI: Registering Knowledge in VLAs”](https://arxiv.org/abs/2610.22178)

In the lab’s RoboTwin Clean tests, oscillating episodes fall from 18.6% without memory to 4.1% with it. Kyutai attributes the repeated back-and-forth motion to perceptual aliasing: two different poses can each look like the next step in a demonstration, leaving a memoryless policy to switch directions. A short history can distinguish between them.

## Results across three robot benchmarks

Kyutai evaluates REKI on LIBERO-Plus, RoboTwin Clean2Rand and RoboCasa365, finetuning the pretrained model separately for each benchmark. LIBERO-Plus perturbs 40 tasks across camera view, robot starting state, language, lighting, background, sensor noise and object layout. RoboTwin Clean2Rand tests models trained on clean scenes under randomized conditions; RoboCasa365 evaluates kitchen tasks across atomic and composite groups.

REKI has the highest out-of-distribution success among models pretrained on public data on LIBERO-Plus and RoboTwin Clean2Rand, despite being the smallest in its comparisons, the lab reports. On RoboCasa365, REKI scores 38.4, compared with 36.0 for RLDX-1 and 35.9 for Qwen-RobotManip; Kyutai identifies both of those systems as pretrained on in-house data. Its published comparison says larger models—roughly twice REKI’s size—are the ones that outperform it on RoboTwin Clean2Rand and RoboCasa365.

The 21.8-millisecond figure is a lab benchmark, not a general measure of how quickly a physical robot completes a task. Kyutai measured one action chunk from raw observations on a single H100, using LIBERO inputs and compiled model code. In that same compiled comparison, the lab lists X-VLA at 25.5 milliseconds, π0.5 at 33.8, ABot-M0 at 46.1 and InternVLA-A1.5 at 51.3.

## Public-data training, released checkpoints

REKI was pretrained for 200,000 steps on 32 H100 GPUs using public robot-manipulation, egocentric-human and vision-language data. The human-manipulation portion includes about 350,000 episodes and 170 million frames. The lab keeps each dataset’s native control convention rather than remapping every source into a shared action space.

The release includes training and evaluation code, data-processing tools, pretrained weights and benchmark-specific finetuned checkpoints. Those checkpoints target several different platforms, including the Franka Panda, ALOHA bimanual robot, mobile PandaOmron and Fourier GR-1 humanoid. Kyutai also describes autonomous rollouts on a UR5 arm, finetuned for one hour on 50 demonstrations per task. The listed checkpoints are tailored to their respective benchmarks and robots, rather than presented as one already-universal policy.

## Sources

1. Original source [Kyutai · page 1](https://kyutai.org/reki/) kyutai.org

Other pages read

2. [Kyutai · page 2](https://kyutai.org/blog/) kyutai.org
3. [arXiv](https://arxiv.org/abs/2610.22178) arxiv.org

## Companies in this story

- [Kyutai](https://ai.info/companies/kyutai)

## Related stories

- [Kyutai’s Voice of Reason Scores 70.6% on GSM8K](https://ai.info/news/kyutai-voice-of-reason-77-1-percent-spoken-math)
- [Artificial Analysis Puts Frontier Models on a Real Robot Arm](https://ai.info/news/artificial-analysis-aa-robotics-real-robot-arm-benchmark)
- [Reka Puts Text, Video and Robot Control in 19B Rho-1](https://ai.info/news/reka-rho-1-19b-omni-model)
- [GPT-6 Astra Tops 40 Robot Policies, but Precision Still Fails](https://ai.info/news/gpt-6-astra-tops-40-robot-policies-but-precision-still-fails)
- [Black Forest Labs Releases 7B FLUX 3 Action Robotics Models](https://ai.info/news/black-forest-labs-flux-3-action-robotics-models)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/kyutai-releases-reki-robot-action-model
