180 models ranked for robotics and physical automation. Scored with heavy bonuses for reasoning (planning & control), vision (sensor processing), function calling (tool use), and JSON mode (structured commands).
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7Anthropic | 95 |
| 2 | GPT-5.5OpenAI | 93 |
| 3 | Gemini 3.1 Pro Preview Custom ToolsGoogle | 92 |
| 4 | Gemini 3.1 Pro PreviewGoogle | 92 |
| 5 | GPT-5.4 ProOpenAI | 92 |
| 6 | GPT-5.4OpenAI | 92 |
| 7 | GPT-5.5 ProOpenAI | 91 |
| 8 | GPT-5.2 ProOpenAI | 91 |
| 9 | Claude Opus 4.6 (Fast)Anthropic | 90 |
| 10 | Claude Opus 4.6Anthropic | 90 |
| 11 | GPT-5.2-CodexOpenAI | 90 |
| 12 | GPT-5.2OpenAI | 90 |
| 13 | Grok 4.20xAI | 89 |
| 14 | GPT-5.3-CodexOpenAI | 89 |
| 15 | GPT-5 ProOpenAI | 89 |
| 16 | Gemini 3 Flash PreviewGoogle | 88 |
| 17 | Grok 4xAI | 88 |
| 18 | GPT-5.1-Codex-MaxOpenAI | 88 |
| 19 | GPT-5 CodexOpenAI | 88 |
| 20 | GPT-5OpenAI | 88 |
| 21 | GPT-5.1OpenAI | 87 |
| 22 | GPT-5.1-CodexOpenAI | 87 |
| 23 | GPT-5.1-Codex-MiniOpenAI | 87 |
| 24 | o3 Deep ResearchOpenAI | 87 |
| 25 | o3 ProOpenAI | 87 |
| 26 | o3OpenAI | 87 |
| 27 | Claude Sonnet 4.6Anthropic | 85 |
| 28 | Claude Opus 4.5Anthropic | 85 |
| 29 | Gemini 2.5 ProGoogle | 84 |
| 30 | Gemini 2.5 Pro Preview 06-05Google | 84 |
Reasoning models decompose complex tasks into action sequences. Function calling enables direct integration with robot control APIs, ROS topics, and hardware interfaces.
Vision-capable models process camera feeds for object detection, scene understanding, and spatial reasoning. Critical for pick-and-place, navigation, and quality inspection.
Generate ROS nodes, control algorithms, sensor fusion pipelines, and simulation configurations. JSON mode produces structured command sequences for robot execution.
Open-source models can run on-device for low-latency inference. Self-hosted options provide the data privacy and deterministic behavior required for safety-critical systems.
Models generate ROS2 nodes, control algorithms, path planning code, and sensor fusion logic. Reasoning handles complex kinematics calculations and state machine design. Vision models analyze sensor data and camera feeds for object detection and navigation.
Models generate C++ and Python control loops, PID implementations, and trajectory planners. For hard real-time requirements, use AI-generated code as a starting point and optimize for deterministic execution. Models understand timing constraints and priority scheduling.
Vision models process camera data for object recognition and scene understanding. Reasoning handles sensor fusion from lidar, IMU, and cameras. Function calling integrates with ROS topics and services. JSON mode outputs structured perception data.
Models generate Gazebo simulation configurations, test scenarios, and evaluation metrics. They write unit tests for control algorithms and integration tests for ROS2 systems. Reasoning helps design edge cases for thorough testing coverage.