SquadSquad

Switching AI Models — Budget vs Quality

Try this to reduce costs temporarily:

Switch everyone to Luna — I'm trying to save costs this sprint

Try this to maximize quality:

Use Sol for all agents — quality over cost for this session

Try this to restore defaults:

Reset to default model strategy

Squad supports 17 models across three tiers. Switch the entire team, set per-agent preferences, or let Squad auto-route based on task type.

1. The Default: Cost-First Strategy

Model selection strategies for different needs. Squad supports 17 models with fallback chains.

Squad’s default model configuration prioritizes cost-effectiveness:

  • Non-code tasks (planning, triage, decisions) → gpt-5.6-luna (fast, cheap)
  • Code tasks (implementation, refactoring) → gpt-5.6-terra (balanced quality/cost)
  • Code review (critical analysis) → gpt-5.6-sol

This is optimized for everyday use — good quality without burning your API budget.

2. Budget Mode: Always Use Luna

If you’re cost-sensitive or working on a small project:

> Switch the entire team to gpt-5.6-luna for all tasks.
✅ Model configuration updated

All agents now use gpt-5.6-luna (fast/cheap tier)

Agents will be faster but may need more guidance on complex tasks.

This is written to .ai-team/model-config.json:

{
  "default": "gpt-5.6-luna",
  "overrides": {}
}

Now every agent uses Luna for everything. Faster responses, lower cost, slightly less nuanced reasoning.

3. Quality Mode: Use Sol for Critical Work

If you’re working on architecture, security, or high-stakes features:

> Use gpt-5.6-sol for Neo (the Lead) and gpt-5.6-terra
> for everyone else.
✅ Model configuration updated

Neo (Lead) → gpt-5.6-sol (premium tier)
All other agents → gpt-5.6-terra (standard tier)

Neo will give higher-quality code reviews and architectural guidance.

This is written to .ai-team/model-config.json:

{
  "default": "gpt-5.6-terra",
  "overrides": {
    "neo": "gpt-5.6-sol"
  }
}

Neo gets the premium model for code reviews and decisions. Everyone else stays on Terra.

4. Per-Agent Overrides

You can set different models for different roles:

> Configure models like this:
> Neo (Lead) → sol (best quality for code review)
> Trinity (Frontend) → terra (balanced)
> Morpheus (Backend) → terra (balanced)
> Tank (Tester) → luna (tests don't need premium reasoning)
✅ Model configuration updated

Neo → gpt-5.6-sol (premium)
Trinity → gpt-5.6-terra (standard)
Morpheus → gpt-5.6-terra (standard)
Tank → gpt-5.6-luna (fast/cheap)

Tank doesn’t need Sol to write tests. Neo does need it for code reviews. Balanced spend.

5. The 17-Model Catalog

Squad supports these models:

Premium tier (highest quality, highest cost):

  • gpt-5.6-sol — premium model for visual and critical work
  • claude-opus-5 — latest Opus, best reasoning
  • claude-opus-4.8
  • claude-opus-4.7
  • claude-opus-4.6

Standard tier (balanced quality/cost):

  • gpt-5.6-terra — current default for code tasks
  • claude-sonnet-5
  • claude-sonnet-4.6
  • claude-sonnet-4.5
  • gpt-5.5
  • gpt-5.4
  • gpt-5.3-codex — OpenAI Codex, code-specialized
  • gemini-3.1-pro

Fast/cheap tier (fast responses, lower cost):

  • gpt-5.6-luna — lightweight model and first fast fallback
  • claude-haiku-4.5 — fast-tier fallback for non-code tasks
  • gpt-5.4-mini
  • gpt-5-mini

Note: Model availability depends on your GitHub Copilot subscription tier.

6. Fallback Chains

If a model is unavailable, Squad falls back to the next tier:

gpt-5.6-luna → claude-haiku-4.5 → gpt-5.4-mini → gpt-5-mini

If Opus is unavailable (rate limit, quota), Squad automatically uses Sonnet. If Sonnet is unavailable, it falls back to Haiku.

You don’t have to configure this — it’s automatic.

7. When to Use Which Model

Use Luna (gpt-5.6-luna) when:

  • Running triage or planning tasks
  • Generating boilerplate code
  • Refactoring (simple renames, restructuring)
  • You’re on a budget and speed matters more than depth

Use Terra (gpt-5.6-terra) when:

  • Writing tests
  • Writing feature code
  • Implementing APIs or UI components
  • Refactoring with logic changes
  • Most everyday development tasks

Use Sol (gpt-5.6-sol) when:

  • Code review (the Lead should catch subtle bugs)
  • Architectural decisions
  • Security-sensitive code
  • Complex debugging
  • Critical features where quality trumps cost

8. Sample Prompts for Model Configuration

Check current model configuration:

> What models is the team using?

Switch everyone to budget mode:

> Switch all agents to luna. We're prototyping, speed matters
> more than perfection.

Use premium for the Lead only:

> Neo should use sol for code reviews. Everyone else stays on terra.

Temporary override for a specific task:

> Morpheus, use sol for this security-critical auth implementation.

Reset to defaults:

> Reset model configuration to Squad's defaults.

Tips

  • Default is fine for most projects. Luna for planning, Terra for code. You don’t need to change it.
  • Use Sol for the Lead. Code reviews benefit most from premium reasoning. Sol catches edge cases Terra misses.
  • Luna is underrated for tests. Test writing doesn’t require deep reasoning — Luna is fast and accurate enough.
  • Per-agent overrides are cheap. Put Sol on the Lead, Luna on the Tester, Terra on everyone else. Balanced budget.
  • Model config is in .ai-team/model-config.json. Commit it so your team uses the same models.
Edit this page on GitHub