Skip to content

Reviewers spawn as Opus 4.7 instead of Sonnet as expected #679

@npwalker

Description

@npwalker

When running under opus 4.6 before 4.7 came out my reviewers would spawn as sonnet 90% of the time.

The skill that launches them says they should be run as mid tier models like sonnet but it seems like since 4.7 that is being followed a lot less than it used to when I analyzed my session history. As a result I'm also burning through usage when reviewer run as opus 4.7 instead of sonnet.

Anyone else running into this issue?

Fwiw, these are the types of answers I get when I ask why the reviewers weren't spawned as sonnet

ME: I see you ran the reviewer agents as opus 4.7 instead of sonnet, even though the skill directs use of a mid tier model. Why did you choose not to explicitly pass sonnet as the model?

CLAUDE: Straight answer: I missed it. The skill text explicitly says "All other persona sub-agents and CE agents use the platform's mid-tier model... In Claude Code, pass model: "sonnet" in the Agent tool call" and names the three exceptions (correctness, security, adversarial). I was focused on scope, reviewer selection, and brief-writing for 12 parallel agents, and didn't re-check Stage 4's model-tiering instruction before dispatch. I should have passed model: "sonnet" to 9 of the 12 Agent calls (everyone except correctness and adversarial — security wasn't dispatched).

CLAUDE: No defensible reason. That's a cost miss: 9 reviews on Opus 4.7 that should have been on Sonnet 4.6. Nothing to do about this run retroactively — the reviews are done — but I'll save the rule to memory so it carries forward.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions