Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation
Kyutai, with Inria and École des Ponts, released REKI, a 2.4B vision-language-action model pretrained on public data with a memory of its recent past for robot manipulation.
Key Takeaways
- Key Highlight：Kyutai, with Inria and École des Ponts, released REKI, a 2.4B vision-language-action model pretrained on public data with a memory of its recent past for robot manipulation.
- Innovation & Tech：Highlights advancements in Meet, Kyutai, REKI, demonstrating rapid progress in model capabilities.
- Industry Impact：Reported via MarkTechPost, offering actionable signals for developers and technology leaders.
REKI is a vision-language-action (VLA) model with 2.4 billion parameters, developed by Kyutai in collaboration with Inria and École des Ponts. Unlike many VLA systems that rely on proprietary datasets, REKI was pretrained exclusively on public data, which lowers the barrier for researchers and practitioners who want to build on it without access to closed corpora.
A distinguishing feature of REKI is its lightweight memory of recent past observations. Rather than treating each manipulation step as an isolated decision, the model carries context from its recent trajectory, which is important for tasks where the current state alone is insufficient to infer intent or progress.
The model was evaluated on out-of-distribution manipulation tasks, where it reportedly performed well on three of the benchmarks tested. Out-of-distribution robustness is a key challenge for embodied AI, since real-world environments rarely match training conditions exactly.
The release matters because it pushes VLA research toward open, publicly pretrained models with explicit memory mechanisms. If the approach generalizes, it could make efficient robot manipulation more accessible to academic and industry teams that lack large-scale proprietary data.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.
Industry Insights & Analysis
As artificial intelligence rapidly evolves, breakthroughs surrounding Meet, Kyutai, REKI, The are shifting toward scalable, robust real-world implementations.
Driven by both open-source ecosystems and proprietary model architectures, the integration between compute optimization, data engineering, and agentic workflows is accelerating. This development provides a strategic benchmark for upcoming AI tooling and developer workflows.