Choosing reviewer speed

In AI Review (one task) or Judge (a task group), select your reviewers and open Model, Effort & Speed. Each provider has its own model and effort settings. Codex also has Standard, Fast, and Ultrafast, where available. The same controls are in the launcher's automatic review settings. If you choose several Codex reviewers there, they all use the displayed Codex settings; other providers keep their own settings.

Reviewer speed defaults to Standard, independently of the implementation task and its remembered launcher speed. Model and effort default to the reviewer's environment/server configuration. Old saved settings keep these defaults. Saved automatic review settings and a launch resumed after sign-in preserve your selection, including when launching a template or objective.

Fast and Ultrafast require existing server/environment permission and a supported Codex model and endpoint. Ultrafast additionally requires server opt-in and a supported explicit or effective model (currently gpt-6-astra); Auto model selection cannot be used for Ultrafast. Incompatible selections fail with an explanation instead of silently switching speed. No policy is enabled by choosing a reviewer speed. Faster modes can consume more allowance or cost more; the control shows the existing account billing guidance.

The review task's model badges show accelerated speed as configured. Later auto-review/verification rounds and judge resubmissions retain the judge's selection and check current policy again. The resubmit dialog also lets you change Codex reviewer speed, including returning to Standard. Sending review feedback does not change the implementation task's speed.

MCP

Single reviewer (create_judge_task):

{
  "taskId": "completed-task-id",
  "agent": "codex",
  "model": "gpt-6-astra",
  "reasoningEffort": "high",
  "speed": "ultrafast"
}

Mixed reviewers use reviewerSettings instead of top-level selection fields:

{
  "groupId": "group-id",
  "agents": ["claude", "codex"],
  "reviewerSettings": {
    "claude": { "reasoningEffort": "high" },
    "codex": { "model": "gpt-6-astra", "reasoningEffort": "high", "speed": "fast" }
  }
}

Only named reviewers may appear in the map. Speed is Codex-only; omit it for other providers. Add reviewLoop to a single-variant review to keep these settings across rounds. To retry safely, supply a UUID requestId and repeat identical arguments. Changing a model, effort, or speed with the same request ID is rejected; a new intentional run needs a new ID.

For the full tool contract, see Judge and review runs.

MCP​

MCP