← Models

Gemini Robotics-ER 1.6

Google · Gemini Robotics

Google DeepMind's robotics-tuned Gemini for embodied reasoning, spatial planning, and physical agent tasks.

Type
multimodal
Context
1M tokens
Max Output
66K tokens
Status
preview
API Access
Yes
License
proprietary
robotics embodied-reasoning vision spatial-reasoning tool-use preview
Released April 2026 · Updated May 1, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on May 1, 2026.

Gemini Robotics-ER 1.6 is Google DeepMind’s April 14, 2026 robotics-tuned Gemini variant. It is built to act as the high-level reasoning model for a robot, calling tools such as Google Search, vision-language-action (VLA) models, or user-defined functions, while handling spatial reasoning, multi-view understanding, and instrument reading natively. DeepMind positions it as a meaningful step over Gemini Robotics-ER 1.5 on practical industrial tasks: instrument reading accuracy reportedly jumps from 23% to 93% on agentic vision evaluations, and spatial reasoning improves on pointing, counting, and success-detection tasks.

This is a specialized model. For typical chat or coding workflows, the standard Gemini 3.1 Pro line is the right reference. Robotics-ER 1.6 is the entry to consult when the question is “how does Google approach embodied reasoning for physical agents?”

Capabilities

DeepMind’s release materials highlight a specific capability profile:

  • High-level embodied reasoning for robots, with native tool calling against VLA models, search, and custom functions.
  • Spatial and physical reasoning improvements on pointing, counting, multi-view understanding, and success detection.
  • Instrument reading at roughly 93% accuracy on agentic-vision tasks, including analog gauges, pressure meters, and sight glasses — a capability essentially absent in the prior 1.5 release.
  • Improved compliance with safety policies on adversarial spatial-reasoning tasks, with DeepMind describing it as the safest robotics model in the family to date.

Robotics-ER 1.6 is already deployed inside Boston Dynamics’ Spot robot for live industrial inspections, which gives the release more practical signal than a typical research preview.

Technical Details

Public anchors at this snapshot:

  • Built on the Gemini multimodal foundation with text, image, and video inputs.
  • Designed as a planner and reasoning model that issues tool calls to lower-level VLA controllers, not as an end-to-end action policy.
  • Available through the Gemini API and Google AI Studio at this snapshot.
  • Marked as a preview release; DeepMind may adjust capabilities and access before GA.

Specific token-window numbers in the frontmatter are best-effort placeholders. Robotics-ER 1.6 inherits Gemini’s standard long-context capacity but exposes a different practical envelope when used in real-time robotics loops.

Pricing & Access

Pricing was not separately published as a standalone Robotics-ER tier at the time of this snapshot. Standard Gemini API rates apply when calling the model through the Gemini API or Vertex AI integrations.

Access options:

  • Gemini API for developers
  • Google AI Studio for prototyping
  • Robotics partners through Google DeepMind’s integration paths
  • Live deployment alongside VLA models such as Gemini Robotics 1.5

Best Use Cases

Choose Gemini Robotics-ER 1.6 for:

  • Industrial inspection workflows where reading analog gauges and instruments matters.
  • Robotics planners that delegate motor control to VLA models while keeping high-level reasoning in a frontier model.
  • Embodied agents that need spatial reasoning, multi-view understanding, and tool-call orchestration in one surface.
  • Research and prototyping work building on Boston Dynamics Spot or similar physical platforms.

This is not the right model for general assistant work, chat, or non-physical agentic coding; the standard Gemini 3.1 Pro and Flash lines are better defaults there.

Comparisons

  • Gemini Robotics-ER 1.5 (Google): Direct predecessor; 1.6 substantially improves instrument reading and spatial reasoning while introducing stronger safety behavior.
  • Gemini 3.0 Flash (Google): General-purpose Gemini Flash variant; Robotics-ER 1.6 is the robotics-specialized counterpart with explicit embodied-reasoning training.
  • Other embodied-AI research models: Typically narrower research demos; Robotics-ER 1.6 differentiates through industrial deployment and multi-view spatial reasoning.