Multimodal foundation model lab targeting enterprises and developers who need vision-language capabilities beyond text-only APIs. Builds the Core, Flash, and Edge model family; the 7B Edge model runs in llama.cpp, uses 3x fewer input tokens than comparable 8B VLMs, and is designed for on-device physical AI systems including robotics and in-vehicle agents.
Reka's open-sourced RIDM reads motion from raw video and outputs keyboard-and-mouse style commands, unlocking action-labelled data for world model training.
Reka open-sources 10,312 hours of unscripted, first-person household video under Apache 2.0, giving physical AI researchers a large-scale real-world training corpus
Reka's WorldModelGym shifts world model evaluation from visual realism to decision-making accuracy, launching an open leaderboard with a simple API submission protocol