Feature Request: Automatic model fallback when quota is exhausted

Feature Request: Automatic Model Fallback on Quota Exhaustion

Background

Claude Code currently offers an opusplan model setting (an existing feature) that intelligently uses:

• Opus for the planning phase (complex reasoning and architecture decisions)
• Sonnet for the execution phase (code generation and implementation)

This is great for optimizing both quality and cost!

Problem

However, if the Opus quota is exhausted during the planning phase itself, the session fails rather than automatically falling back to Sonnet for planning as well.

This creates challenges in enterprise/multi-user environments where Opus quota needs to be managed carefully across multiple users.

Current Behavior

• opusplan uses Opus for planning, Sonnet for execution
• If Opus quota runs out during planning → session fails ❌
• User must manually switch to Sonnet using /model sonnet

Desired Behavior

Extend the existing model configuration to support automatic fallback when quota is exhausted:

{
  "model": "opusplan",
  "modelFallback": "sonnet"
}

Or more generally for any model:

{
  "model": "opus",
  "modelFallback": {
    "onQuotaExhausted": "sonnet",
    "onRateLimit": "sonnet"
  }
}

When Opus quota is exhausted:

1. Gracefully switch to the fallback model (Sonnet) ✅
2. Show a notification to the user about the fallback
3. Continue the session without interruption

Use Cases

1. Enterprise settings: IT administrators can configure system-wide settings that ensure continuous operation even when Opus quota runs out
2. Multi-user environments: Teams sharing quota can have graceful degradation
3. Cost management: Automatically manage costs by falling back to less expensive models
4. Unattended operations: Long-running tasks won't fail mid-session

Benefits

• Improved user experience (no session interruptions)
• Better quota management for teams
• Cost optimization
• Aligns with the existing opusplan philosophy of intelligent model selection

Alternatives Considered

• Setting ANTHROPIC_SMALL_FAST_MODEL environment variable, but this doesn't provide automatic fallback on quota exhaustion
• Manually switching with /model sonnet during sessions (not practical for unattended operations)

---

Thank you for considering this enhancement to the existing opusplan feature!