[FEATURE] Per-tool model configuration

Resolved 💬 2 comments Opened Jan 24, 2026 by cr4b5 Closed Mar 1, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Feature Request: Per-tool model configuration

Summary

Allow users to manually specify which model should be used for each tool in Claude Code (e.g., Haiku for read_file, grep, list_directory; Sonnet/Opus for write_file, edit_file).

Motivation

Claude Code uses various tools under the hood (file reading, writing, shell commands, etc.). These tools have vastly different complexity requirements:

  • Reading a file doesn't require advanced reasoning — any model can do it
  • Editing code or writing new files benefits from the most capable model available

Currently, one model handles everything. Letting users tag each tool with a preferred model would give fine-grained control over the cost/performance tradeoff.

Proposed Solution

Proposed Solution

A configuration option where users explicitly map tools to models:

{
  "toolModels": {
    "read_file": "claude-haiku-4-5-20251001",
    "list_directory": "claude-haiku-4-5-20251001",
    "grep": "claude-haiku-4-5-20251001",
    "bash": "claude-sonnet-4-5-20250929",
    "write_file": "claude-opus-4-5-20251101",
    "edit_file": "claude-opus-4-5-20251101",
    "default": "claude-sonnet-4-5-20250929"
  }
}

Users would have full control — no magic or automatic detection, just explicit mappings.

Alternative Solutions

Additional Context

This could start as a simple rule-based system (certain commands → Haiku) and evolve into smarter automatic detection over time.`,

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

Benefits

  • Transparent: Users know exactly which model runs which tool
  • Customizable: Different users can optimize differently based on their priorities (cost vs. speed vs. quality)
  • Simple to implement: No complexity detection needed — just a lookup table
  • Predictable: Behavior is deterministic based on config

Additional Context

Additional Context

This approach keeps the user in control rather than relying on heuristics. Power users who want to optimize their Claude Code usage would appreciate this level of granularity.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗