# Claude Code delegate small tasks to local models — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/b0459a877ae67825

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-10-21 03:08 ingested 2026-10-22 02:42

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xbc3r0/claude_code_delegate_small_tasks_to_local_models/)

## retained text



```
Has anyone experimented with letting Claude code delegate smaller tasks to local models to save tokens?

My setup:

* My laptop run a local model (qwen3.6-35b-a3b, 32k token limit) with LMStudio and exposed its server to tailscale network
* My pc use claude code
* Both use yeschef-cli to let claude code delegate its tasks to my local model

I tried 2 approach:

* Claude code plan, ochestrate and let local model write all the code, then claude code review. The downside of this is it's very slow (depends on my laptop)
* Claude code let my local model write small helper functions and test them itself before sending back to claude code. The downside of this is there's too many tool calls, and local model is not very smart so the tool calls keep failing

Both approaches also use more tokens that Claude Code Opus 5.5 just doing it alone. Because the things that Claude Code now do instead - writing the prompts, mcp calls and reviewing - already outweighs the output token local model save.

Any recommendation?
```



document fetched 2026-10-22 00:44 · canonical https://reddit.com/r/ClaudeCode/comments/1xbc3r0/claude_code_delegate_small_tasks_to_local_models
