model-cost-compare

Classify the task first: mechanical work like renames goes to included-cost or budget seats, standard coding fits mid-tier models, and hard reasoning justifies premium seats. Then estimate token volume from files touched and compute cost per model using the price table.

Estimate input and output tokens from the task scope, such as files touched multiplied by average size, then multiply by each model's per-million-token input and output rates. State your volume assumptions explicitly so the estimate is auditable.

Escalate to a premium seat like Opus 5 when the task touches API or schema contracts, security-sensitive code, CI configuration, release artifacts, user-facing UI, new modules, or breaking changes. Cheap-seat agreement never settles judgment-class decisions.

Not always. Fable 5 is explicitly excluded from security audits because its safety classifiers can refuse offensive-security phrasing; security review goes to Opus 5 instead. Price alone does not determine fitness for sensitive task types.

Estimates depend on rough token volume assumptions, so actual costs vary with real context sizes and retry counts. The skill flags any estimate exceeding $1 before dispatch, but it cannot predict exact billing for open-ended tasks.