Back to rankings

Claude Code Subagents and Teammates on a Different Model Now Use That Model's Own Effort Level

Claude CodeView original changelog

Claude Code 2.1.317 changes how effort is chosen for subagents and agent teammates that run on a different model from the main conversation. They now use that model's saved effort level, or its default, instead of inheriting the session's effort level.

Key Takeaways

  • Effort is now per model for subagents and teammates that run on a different model than the main session.
  • The session's effort no longer leaks into delegated work on another model.
  • A model's saved effort level wins, with the model's default used when nothing is saved.
  • Mixed-model agent teams are the most affected, since lighter teammates stop inheriting a high lead-model setting.
  • Cost and latency may shift, so users who relied on inheritance should review their saved per-model effort levels.
  • Same-model subagents are unaffected, since the change applies only when the model differs from the main conversation.

What changed

In Claude Code 2.1.317, a subagent or agent teammate configured to run on a different model from the main conversation no longer inherits the session's effort level. It uses the effort level saved for its own model, or that model's default if none is saved.

Why it matters

Effort levels are not interchangeable across models. A setting such as xhigh chosen for a flagship model in the main session could previously carry over to a cheaper or faster model used for a delegated task. That could mean unexpected cost, slower runs, or an effort value the other model handles differently. With 2.1.317, each model keeps its own effort preference, so a user who tuned a smaller model for quick lookups sees that tuning respected when it is spawned as a subagent.

What users should check

Anyone who relied on subagents mirroring the main session's effort should review the effort level saved for each model they delegate to. Teams that run mixed-model agent teams, with a strong lead model and lighter teammates, are the most likely to notice the change in cost and latency.