Reka AI

Models that work across media

Reka AI, branded Reka, develops models and infrastructure for interpreting text, images, audio and video. Its current work emphasizes video understanding and AI connected to real-world data. The website's terms identify the company as Reka AI, Inc. [1][2]

The company introduced Reka Core on April 15, 2024. Core was a model trained from scratch to accept multiple input modalities, with a stated 128,000-token context window. Processing a video or an audio recording alongside text made the release relevant to applications beyond text-only chat. [3]

Model access and evidence

Reka offered the first Core version through its service and described enterprise deployment options. The announcement established model availability, without releasing downloadable weights. Its comparisons with competing models were Reka's reported evaluations, rather than an independent demonstration that Core performed equally well on every task. [3]