共有:
Merger · Physical AI

Reka merges with Moonvalley——
ex-DeepMind Veo researchers on board

Reka has merged with Moonvalley, bringing in a research team of alumni from Google DeepMind, Meta, and Google — including contributors to Veo, DeepMind's video generation model. Physical AI model development accelerates overnight.

AI Navigate Editorial·2026.06.13·6 min read
Reka Multimodal LLM lab Moonvalley Video gen · DeepMind alumni Merged
01
Why This Merger

Reka lacked the research
depth for physical AI

Reka had earned recognition as a capable multimodal LLM lab handling text, images, and video, but remained lightly staffed for the hardware-intensive push into physical AI — robotics, sensor integration, and real-world environment understanding. Moonvalley specialised in video generation and came staffed with DeepMind alumni who had worked on cutting-edge video models.

Reka merged with Moonvalley (June 2026). The deal brings in a research team including Veo contributors and alumni from DeepMind, Meta, and Google, accelerating physical AI model development.


02
Who Joined

Veo experience and
physical-AI pedigree

Moonvalley's research team spans video generation, physical simulation, and multimodal foundations — three fields that together underpin physical AI.

DeepMind alumni Veo experience Video gen foundations Meta alumni Multimodal scale Large model training Google alumni Sensor fusion Physical AI systems
FIG. Moonvalley's research team by origin and primary expertise area.

Researchers with Veo experience bring direct know-how in generating physically plausible video — a capability that transfers naturally to robotic perception and environment modelling. Combined with Reka's existing multimodal LLM base, the merged lab gains a credible physical-AI research stack.

03
Who This Matters To

Mostly a watch-and-wait story
outside robotics and sensors

Reka's future models are likely to lean more heavily into physical AI, but text and image processing use cases are not affected for now. If you're working in robotics, autonomous systems, or sensor-rich domains, Reka's roadmap is worth tracking more closely. For general LLM workloads, the near-term impact is minimal.

AI Navigate — Daily Update · 2026.06.13