GitHub Copilot: Model Selection for Claude and Codex Third-Party Agents
GitHub has expanded model selection capabilities to the Claude and Codex third-party coding agents on github.com, allowing developers to choose a specific underlying model when kicking off a task. For the Claude agent, available models include Claude Sonnet 4.6, Claude Opus 4.6, Claude Sonnet 4.5, and Claude Opus 4.5; for Codex, developers can select from GPT-5.2-Codex, GPT-5.3-Codex, and GPT-5.4. The feature brings parity with the existing Copilot cloud agent's model picker and is included in all existing paid Copilot subscriptions (Business and Enterprise require administrator enablement of the relevant agent policy).
Sources & Mentions
4 external resources covering this update
Model Selection Comes to Third-Party Coding Agents
GitHub has extended its model selection feature to the Claude and Codex third-party coding agents available on github.com, closing a capability gap that previously existed between the native Copilot cloud agent and its third-party counterparts. Developers can now choose which underlying AI model powers their coding agent session before kicking off a task — giving them meaningful control over the trade-off between speed, cost, and reasoning depth.
Available Models by Agent
Claude Agent
The Claude agent now surfaces four model options at task initiation:
- Claude Sonnet 4.6 — the latest and most capable Sonnet variant, balancing speed with strong reasoning
- Claude Opus 4.6 — the most powerful Claude model, suited for complex multi-step coding tasks
- Claude Sonnet 4.5 — a prior-generation Sonnet for teams with established workflows
- Claude Opus 4.5 — the previous flagship Opus model, still available for continuity
Codex Agent
The Codex agent offers three model options:
- GPT-5.2-Codex — the baseline Codex model optimized for code generation
- GPT-5.3-Codex — a mid-tier option offering improved accuracy on complex tasks
- GPT-5.4 — the most advanced Codex model, delivering the highest-quality outputs for demanding engineering tasks
How It Works
When assigning a task to either the Claude or Codex agent through the GitHub Copilot interface on github.com, a model picker appears alongside the task configuration. Developers select their preferred model from the dropdown before the agent begins work. The selected model applies to the entire agent session for that task.
This experience mirrors the model selection already available for the native GitHub Copilot cloud agent, bringing the third-party agents to full feature parity on this dimension.
Access and Subscription Requirements
Model selection for the Claude and Codex agents is available at no additional cost — it is included in all existing paid GitHub Copilot subscriptions:
- Copilot Individual / Pro — available directly
- Copilot Business — requires an administrator to enable the Claude and/or Codex agent policy in organization settings
- Copilot Enterprise — same administrator enablement requirement as Business
The feature is currently in public preview on github.com. Organizations that have not yet enabled third-party agent policies will need to do so before their developers can access this capability.