AI Models & Tokens

SKYCOT uses Anthropic's Claude Opus and Sonnet models with adaptive routing that balances quality and cost automatically.

Adaptive Model Routing

Not every build step requires the same level of intelligence. SKYCOT routes each session to the most appropriate model based on the task complexity. High-stakes stages like compilation and architecture run on Opus for maximum accuracy. Parallel tasks like UI generation and test writing run on Sonnet for speed and cost efficiency. This achieves roughly 57% cost reduction compared to using Opus for every session.

Session TypeModelEffortTier Override
Intent extraction + compilationOpusMaxAlways Opus
Foundation (schema, config)OpusMaxAlways Opus
Integration sessionsOpusHighAlways Opus
Independent featuresSonnetMediumPro+ can request Opus
UI componentsSonnetMediumPro+ can request Opus
Test generationSonnetMediumNo override
Code reviewSonnetLowBusiness gets Opus review
VerificationSonnetLowNo override
Recovery sessionsOpusHighAlways Opus

Token Tracking

Every API call to Claude is metered and logged in real-time. Your account dashboard shows a breakdown of token usage by project, by build, and by session type. You always know exactly how many tokens remain on your plan and where they went.

Opus tokens cost more than Sonnet tokens, so the routing table above directly impacts your per-build cost. Higher-tier plans provide more token headroom and lower overage rates.

Pre-Build Cost Estimation

Before any build starts, SKYCOT displays a transparent cost estimate. A complexity scorer analyses your selected archetype, features, and follow-up answers to produce a complexity score from 1 to 10. The cost estimator then maps this score to predicted token usage per session and calculates the dollar cost at your plan's rate.

Estimates include a confidence interval (for example, "$3.20 - $5.50, most likely $4.20") so you can decide whether to proceed. After the build completes, SKYCOT shows actual vs estimated cost so you can gauge accuracy over time.

Typical cost ranges (to you, at plan rates):

  • Simple landing page: $0.50 - $1.00 (mostly Sonnet)
  • Medium CRUD app: $2 - $4 (hybrid routing)
  • Complex SaaS: $5 - $10 (Opus-heavy)
  • Custom research + build: $10 - $35 (Opus-heavy, multi-hour research phase)

SKYCOT uses prompt caching to reduce costs by up to 90% on repeated system prompts and archetype definitions.

Plan Token Allowances

PlanPriceTokens/moBuildsModelsOverage
Starter$19/mo2M~3-5 simpleSonnet only$12/M
Pro$49/mo10M~5-8 mediumSonnet + Opus (compilation)$8/M
Business$99/mo25M~10-15 anyFull Opus access$5/M
EnterpriseCustomCustomUnlimitedCustomCustom

Build counts are estimates based on typical application complexity. Actual usage depends on archetype, feature count, and model routing. See FAQ for details on overage billing.