Gemini Robotics 2

Gemini Robotics 2 is the second generation of Google DeepMind's physical-AI stack, released on 30 July 2026 as three models rather than one. Gemini Robotics ER 2 is the reasoning layer, built on Gemini 3.5 Flash: it takes interleaved text, images, video and audio, works within a 128,000-token context and returns up to 64,000 tokens of plans, spatial references and pointing. Its headline change over the ER 1.x line is temporal — instead of judging a static frame or a short clip, it reasons over a continuous video stream and keeps track of what it has already done across minutes of work. Gemini Robotics 2 is the vision-language-action controller that turns those plans into whole-body motion, and Gemini Robotics On-Device 2 runs on the robot itself, so a machine can keep working without a network connection. The reasoning model is generally available through the Gemini API and Google AI Studio, with the Gemini Enterprise Agent Platform in private preview; the action models reach robot makers through partnerships rather than open download, and the weights stay closed. DeepMind's own model card is explicit that the system must not be put in charge of safety-critical work — healthcare, transport and anything where a malfunction hurts someone — and reports separate evaluations for safety instruction following and human-safety monitoring. The catalogue keeps the 2025 first generation as a separate entry; this profile covers the 2026 line.

▸News

▸Videos

No videos yet.

▸Reviews