Pending benchmarksTracking since Oct 18, 2026
Is REKI state of the art?Pending benchmarks.Announced Oct 16, 2026; no independent results yet.
Kyutai's 2.4B vision-language-action robot model, built with Inria and École des Ponts, pretrained on public data only, with a built-in memory of its recent past.
Announced facts
The details so far
- Lab
- Kyutai
- Announced
- Oct 16, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- 2.4B (2.2B VLM, 0.2B action expert)
- Type
- Vision-language-action (VLA) robot policy
- Built with
- Inria and École nationale des ponts et chaussées
- Memory
- 32 register tokens per frame give the policy access to its recent past
- Training data
- Public data only
- Weights
- Hugging Face (base and four finetuned checkpoints)
- Code
- Training and inference code to be fully open sourced
- License
- CC BY-NC-SA 4.0 weights, MIT codeunsure · Per Kyutai's Hugging Face model card; Kyutai's X posts don't state a license.
- API
- None stated
Tags marked unsure are not stated on the lab's own pages.
Next steps
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
Self-reported
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: rivals larger models trained on in-house data with complex multi-stage pipelines, setting a new Pareto frontier between performance and size.
Receipts
Sources
LABKyutai announcement on Xx.com/kyutai_labs/status/2111125878709088487 · Oct 16, 2026LABKyutai weights and code post on Xx.com/kyutai_labs/status/2111125887101878462 · Oct 16, 2026REPOreki-2b on Hugging Facehuggingface.co/kyutai/reki-2b · Oct 16, 2026Tracking continues until an independent board lists it.