Skip to content

chore(agents): tier subagent models by judgment vs execution - #225

Merged
zaxovaiko merged 1 commit into
devfrom
chore/BF-0/subagent-model-tiers
Sep 25, 2026
Merged

zaxovaiko merged 1 commit into
devfrom
chore/BF-0/subagent-model-tiers

Conversation

@zaxovaiko

Copy link
Copy Markdown
Member

Summary

  • dev, module-author, plugin-author, contract-reviewer and operator move to opus, joining security-reviewer and expert.
  • qa, quality-reviewer, cleaner and explore stay on sonnet, docs stays on haiku.
  • The consumer template moves builder and debugger to opus, consumers pick it up on the next @openora/create bump.

Why

Opus goes where a miss is costly - money-path implementation, module and plugin authoring, breaking-change and security review, root-cause debugging. Well-specified execution and pattern-matching work (tests, convention review, cleanup, search) stays on sonnet, which is cheaper per token for the same context.

Alternatives considered

Keeping dev and builder on sonnet. It is cheaper, but they are the main implementers on the money path, so rework from a weaker first pass would eat the saving.

Risks

Higher cost per run for the five openora agents plus builder and debugger downstream. No code changes.

Opus for implementers, authors and contract/security review where a miss is costly. Sonnet stays on qa, quality review and cleanup. The consumer template moves builder and debugger to opus.
@zaxovaiko zaxovaiko self-assigned this Sep 25, 2026
@zaxovaiko
zaxovaiko merged commit 0dd1f06 into dev Sep 25, 2026
2 checks passed
@zaxovaiko
zaxovaiko deleted the chore/BF-0/subagent-model-tiers branch September 25, 2026 20:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant