Gemini Robotics ER 2
2026-07-30 · Publisher : Google · Domain : Artificial Intelligence
Google DeepMind launches Gemini Robotics ER 2, its most capable embodied reasoning model, which tracks task progress in continuous video and orchestrates several robots through the Live API.
What changes
- Progress tracking through a continuous video stream: the robot checks that a step is complete before moving on to the next one.
- Precise moment search in a video: 91.3% accuracy and 0.96 s mean absolute distance.
- Multi-robot collaboration with shared semantic understanding.
- Success or failure detection on raw video streams, broader instrument reading (10 types).
- Presented as Google's safest robotics model, with a new safety benchmark.
### Specifications - Availability: Gemini API and AI Studio (public), private preview on Gemini Enterprise Agent Platform. - Identifiers: gemini-robotics-er-2-preview and gemini-robotics-er-2-streaming-preview. - Native tool calling (including Google Search) and custom functions; integrated with the Live API. ### Pricing | Item | Standard (USD / 1M tokens) | Batch | |---|---|---| | Input until 12/31/2026 | 1.00 | 0.50 | | Output until 12/31/2026 | 5.00 | 2.50 | From January 1, 2027: 2.00 and 10.00 (batch 1.00 and 5.00). Pricing recorded from the current pricing page. ### Benchmarks Scores published by the vendor: 91.3% accuracy and 0.96 s mean absolute distance in moment search. The other comparisons (ER 1.6, frontier models) are charts without figures in the text. ### Key takeaways A high-level brain model for robots that delegates motor execution to a vision-language-action model; useful for prototyping orchestrated robot fleets.
Links
Tags : LLM, Google, Gemini, Robotique, Raisonnement incarné