Models

ShipCode can assign a provider, model, and reasoning effort independently to planning, review, execution, and verification. App settings establish the defaults; project and issue overrides can replace them for a narrower scope.

The source of truth is model-catalog.ts, model-resolution.ts, and model-config-presets.ts.

Providers and models

Provider	Curated models	Plan	Review	Execute	Verify
Claude CLI	`opus`, `sonnet`, `haiku`, `fable`, `claude-fable-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001`	✓	✓	✓	✓
Codex CLI	`gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`	✓	✓	✓	✓
Gemini CLI	`gemini-2.5-pro`, `gemini-2.5-flash`	✓	✓	✓	✓
Cursor CLI	`auto`	—	—	✓	—
Grok CLI	`grok-4.5`	—	—	✓	—
OpenRouter	`openrouter/auto`, `openrouter/free`, `anthropic/claude-sonnet-4.6`, `anthropic/claude-opus-4.8`, `anthropic/claude-fable-5`, `openai/gpt-5.6-sol`, `openai/gpt-5.6-terra`, `openai/gpt-5.6-luna`, `qwen/qwen3.6-plus`	✓	✓	✓	✓

Codex publishes its installed model catalog through codex debug models; the models above are ShipCode’s conservative fallback options when an older CLI cannot report capabilities. Cursor and Grok are execute-only because their headless CLIs cannot guarantee a read-only planning, review, or verification run.

The curated OpenRouter list is only a starting point. OpenRouter settings accept other catalog model IDs, and openrouter/auto can choose the upstream model per request. See OpenRouter for authentication and fallback behavior.

Pinned defaults

Pinned defaults are used when a CLI-backed phase or auxiliary task has no explicit model override. They can lag newly added catalog entries deliberately while fleet compatibility is verified.

Lane	Provider	Pinned model
Pipeline phases	Claude	`claude-fable-5`
PRD rewrite	Claude	`claude-sonnet-4-6`
Issue triage	Claude	`claude-haiku-4-5-20251001`
Pipeline phases	Codex	`gpt-5.6-sol`
PRD rewrite	Codex	`gpt-5.6-luna`
Issue triage	Codex	`gpt-5.6-luna`
Pipeline phases	Gemini	`gemini-2.5-pro`
Execution	Cursor	`auto`
Execution	Grok	`grok-4.5`
Paid routing	OpenRouter	`openrouter/auto`
Free routing	OpenRouter	`openrouter/free`
Explicit fallback	OpenRouter	`qwen/qwen3.6-plus`

The phase provider itself comes from app settings, then project overrides, then issue overrides. A model ID follows the same narrowing order. Use the desktop Settings → Pipeline panel for app defaults and project or issue settings for targeted overrides.

Slug aliases

Model fields accept canonical IDs or these case-insensitive shorthands. Bare Claude family aliases are rolling Claude CLI selections: the installed Claude CLI chooses the latest generation in that family. Versioned shorthands resolve to concrete IDs and remain pinned.

Family	Aliases	Resolves to
Opus (latest)	`opus`	Rolling `opus` alias
Opus 4.8	`opus-4.8`, `opus4.8`	`claude-opus-4-8`
Opus 4.7	`opus-4.7`	`claude-opus-4-7`
Opus 4.6	`opus-4.6`	`claude-opus-4-6`
Sonnet (latest)	`sonnet`	Rolling `sonnet` alias
Sonnet 4.6	`sonnet-4.6`	`claude-sonnet-4-6`
Fable (latest)	`fable`	Rolling `fable` alias
Fable 5	`fable-5`, `fable5`	`claude-fable-5`
Haiku (latest)	`haiku`	Rolling `haiku` alias
Haiku 4.5	`haiku-4.5`	`claude-haiku-4-5-20251001`
Grok	`grok`, `grok-4.5`, `grok4.5`	`grok-4.5`
GPT-5.6 Sol	`5.6`, `gpt-5.6`, `5.6-sol`, `sol`	`gpt-5.6-sol`
GPT-5.6 Terra	`5.6-terra`, `terra`	`gpt-5.6-terra`
GPT-5.6 Luna	`5.6-luna`, `luna`	`gpt-5.6-luna`
GPT-5.5	`5.5`	`gpt-5.5`
GPT-5.4	`5.4`	`gpt-5.4`
GPT-5.4 Mini	`5.4-mini`, `mini`	`gpt-5.4-mini`

Unknown values are preserved as typed, which allows provider-native model IDs that are not in the curated fallback catalog.

Rolling Claude aliases are valid only when the selected provider is Claude CLI. OpenRouter requires a concrete OpenRouter model ID such as anthropic/claude-opus-4.8; ShipCode rejects bare Claude aliases before an OpenRouter request can be sent.

ShipCode records the concrete model reported by successful Claude stream-JSON runs. Interactive Claude sessions do not always expose that metadata. When they do not, the resolved-model field remains unavailable rather than recording the requested alias as if it were concrete evidence.

Existing saved concrete Claude IDs are not migrated to rolling aliases. A saved claude-opus-4-8 selection stays pinned until it is explicitly changed.

Presets

Preset	Scope	Plan	Review	Execute	Verify
`claude`	App or project	Fable 5	Fable 5	Fable 5	Fable 5
`codex`	App or project	GPT-5.6 Sol	GPT-5.6 Sol	GPT-5.6 Sol	GPT-5.6 Sol
`hybrid`	App or project	Fable 5	GPT-5.6 Sol	GPT-5.6 Sol	GPT-5.6 Sol
`opus-combo`	Project only	Opus 4.8	GPT-5.6 Sol	GPT-5.6 Sol	GPT-5.4 Mini
`fable-combo`	Project only	Fable 5	GPT-5.6 Sol	GPT-5.6 Terra	GPT-5.6 Luna

opus-combo and fable-combo are project-only because their concrete per-phase model IDs require project overrides; the app-wide preset path selects providers while keeping the pinned provider models. The Fable combo intentionally maps the deepest planning work to Fable, flagship review to Sol, sustained implementation to Terra, and fast verification to Luna.

Reasoning effort

ShipCode normalizes the shared none, minimal, low, medium, high, and xhigh choices to each provider’s supported controls.

Provider or model	Effective choices	Normalization rules
Claude, except Fable 5	None, Medium, High	Minimal/Low → None; Extra high → High
Fable 5 on Claude CLI	Medium, High	None/Minimal/Low → Medium; Extra high → High
Codex	Low, Medium, High, Extra high	None/Minimal → Low
Gemini	Low, Medium, High	None/Minimal → Low; Extra high → High
Cursor	None	The CLI selects reasoning automatically
Grok	None	The CLI selects reasoning automatically
OpenRouter Claude Sonnet 4.6 / Opus 4.6 / Opus 4.8	None, High	Any non-None value uses adaptive thinking
`anthropic/claude-fable-5` on OpenRouter	Low, Medium, High, Extra high	None/Minimal → Low
Other OpenRouter models	Provider-dependent	The requested effort may be remapped upstream

Fable 5 always uses adaptive thinking, so neither path offers a no-thinking mode for it: the Claude CLI exposes Medium and High budgets, and OpenRouter reports the model as mandatory-reasoning with Low through Extra high available, so a None selection is lifted to Low rather than switching reasoning off. A configured effort is normalized before the provider runs, and the effective value is what ShipCode records and displays.