Reka is an AI research lab whose work centers on natively multimodal models it calls Omni models. Rather than bolting separate vision or speech modules onto a text model, Reka designs a single architecture that processes — and in some cases generates — text, image, video, and audio together. The public model families referenced by the company include Spark, Edge, Flash, and Core, positioned for different capability and efficiency points.
The company pairs its models with a deployment and tooling stack. Reka Cloud provides infrastructure for training, customizing, and running models at scale. A dedicated video infrastructure offering is built to tag, reason over, search, and clip large volumes of video, accessible through an API, MCP (Model Context Protocol), or Reka's own app. Claru is a data platform supplying egocentric video, robotics trajectories, world-model footage, and expert human judgment for teams training frontier models.
Reka positions its technology for the physical-AI era and for enterprise use in security, media, and defense. A recurring engineering theme on the site is real-time performance: distillation, quantization, and reinforcement learning are used to keep large omni models within practical latency budgets, such as running inside a 24 frames-per-second video budget on existing hardware. Reka does not publish public per-seat prices; interested organizations are directed to request a demo.
