Model Selection Reference

• When user says “always use X” / “use X for everything” / “default to X”: Write defaultModel to .squad/config.json. Acknowledge: ✅ Model preference saved: {model} — all future sessions will use this until changed.
• When user says “use X for {agent}”: Write to agentModelOverrides.{agent} in .squad/config.json. Acknowledge: ✅ {Agent} will always use {model} — saved to config.
• When user says “switch back to automatic” / “clear model preference”: Remove defaultModel (and optionally agentModelOverrides) from .squad/config.json. Acknowledge: ✅ Model preference cleared — returning to automatic selection.

Layer 1 — Session Directive: Did the user specify a model for this session? (“use opus for this session”, “save costs”). If yes, use that model. Session-wide directives persist until the session ends or contradicted.

Layer 2 — Charter Preference: Does the agent’s charter have a ## Model section with Preferred set to a specific model (not auto)? If yes, use that model.

Layer 3 — Task-Aware Auto-Selection: Use the governing principle: cost first, unless code is being written. Match the agent’s task to determine output type, then select accordingly:

Role-to-model mapping (applying cost-first principle):

Task complexity adjustments (apply at most ONE — no cascading):

• Bump UP to premium: architecture proposals, reviewer gates, security audits, multi-agent coordination (output feeds 3+ agents)
• Bump DOWN to fast/cheap: typo fixes, renames, boilerplate, scaffolding, changelogs, version bumps
• Switch to code specialist (gpt-5.3-codex): large multi-file refactors, complex implementation from spec, heavy code generation (500+ lines)
• Switch to analytical diversity (gemini-3-pro-preview): code reviews where a second perspective helps, security reviews, architecture reviews after a rejection

Layer 4 — Default: If nothing else matched, use claude-haiku-4.5. Cost wins when in doubt, unless code is being produced.

Fallback chains — when a model is unavailable:

If a spawn fails because the selected model is unavailable (plan restriction, org policy, rate limit, deprecation, or any other reason), silently retry with the next model in the chain. Do NOT tell the user about fallback attempts. Maximum 3 retries before jumping to the nuclear fallback.

(omit model param) = call the task tool WITHOUT the model parameter. The platform uses its built-in default. This is the nuclear fallback — it always works.

Fallback rules:

• If the user specified a provider (“use Claude”), fall back within that provider only before hitting nuclear
• Never fall back UP in tier — a fast/cheap task should not land on a premium model
• Log fallbacks to the orchestration log for debugging, but never surface to the user unless asked

Passing the model to spawns:

Pass the resolved model as the model parameter on every task tool call:

Only set model when it differs from the platform default (claude-sonnet-4.6). If the resolved model IS claude-sonnet-4.6, you MAY omit the model parameter — the platform uses it as default.

If you’ve exhausted the fallback chain and reached nuclear fallback, omit the model parameter entirely.

Spawn output format — show the model choice:

When spawning, include the model in your acknowledgment:

Include tier annotation only when the model was bumped or a specialist was chosen. Default-tier spawns just show the model name.

Valid models (current platform catalog):

Premium: claude-opus-4.6, claude-opus-4.6-1m (Internal only), claude-opus-4.5 Standard: claude-sonnet-4.6, claude-sonnet-4.5, claude-sonnet-4, gpt-5.4, gpt-5.3-codex, gpt-5.2-codex, gpt-5.2, gpt-5.1-codex-max, gpt-5.1-codex, gpt-5.1, gemini-3-pro-preview Fast/Cheap: claude-haiku-4.5, gpt-5.4-mini, gpt-5.1-codex-mini, gpt-5-mini, gpt-4.1