Models

ACP supports model choice at the agent and execution layer. Different models are suited to different types of work, and ACP exposes them directly in agents, teams, and thread execution.

Bring your own models

Team and Enterprise plans can bring their own models

ACP supports custom inference endpoints for collaborative plans. Team and Enterprise workspaces can connect OpenAI-compatible endpoints and expose those models inside the rest of the platform without storing credentials in the browser.

Execution models

Claude Opus 4.6
claude-opus-4-6

Highest-capability execution model for complex reasoning, architecture work, and deep coding tasks.

Claude Sonnet 4.5
claude-sonnet-4-5

Balanced flagship model for most ACP agent execution and default coding workflows.

Claude Haiku 4.5
claude-haiku-4-5

Fast and efficient model for lightweight execution, quick iteration, and high-volume routine work.

Gemini 3 Flash
gemini-3-flash

Fast long-context execution model available in ACP product surfaces for broader agent workflows.

Gemini 3.1 Pro
gemini-3-1-pro

Higher-depth long-context model for deeper planning and more involved multi-step reasoning.

Deep research models

Gemini 3 Flash Preview
gemini-3-flash-preview

Default deep research model for faster research passes and broad web synthesis.

Gemini 3 Pro Preview
gemini-3-pro-preview

Higher-capability deep research model for heavier synthesis and long-context research workflows.