REKI Models by Kyutai
Kyutai releases · 1 in the last 12 monthsChangelog →
1 model2026
Last refreshed 2026-10-22. Next refresh: weekly.
Capabilities
VisionAll models
MultimodalAll models
About
Kyutai's REKI vision-language-action models for robot control: a Qwen3.5-2B vision-language model followed by a flow-matching action expert with cached register-token memory, pretrained on public data only.
Decision facts
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
REKI 2BCurrent
Use when the workload needs robotics, 2.4B parameters, and multimodal inputs.
2026-10robotics2.4B parametersmultimodal inputs
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| REKI 2B | Use when the workload needs robotics, 2.4B parameters, and multimodal inputs. | 2026-10 | robotics2.4B parametersmultimodal inputs | Current |
Release Timeline
1 release group2026-10
1 current
REKI 2B
Currentrobotics2.4B parametersmultimodal inputs
Specifications(1 models)
| Model | Released | Parameters | Vision | Multimodal |
|---|---|---|---|---|
| REKI 2B | 2026-10 | 2.4B | Yes | Yes |