Every Forge build is a real Anthropic API call. Sonnet 4.5 costs us $3 per million input tokens and $15 per million output tokens. Opus 4.7 — the deeper-reasoning model behind premium builds — is $5 / $25. A real custom platform build outside the Forge runs $250K to $1.5M+ all-in over six to eighteen months. We do it in 60–90 minutes because Claude is doing the building, a 100-mind council is critiquing it, and the studio has shipped this loop enough times that the patterns are baked in.
Credits are the denomination. 1 credit = $0.01 USD. A simple landing page costs ~5,000 credits ($50). A standard build runs ~15,000 credits ($150). A ReimburseOS-class platform on Opus can hit ~200,000 credits ($2,000) of build power. Each subscription tier comes with a monthly credit grant; when you spend past it, credit packs top you up at a tier-discounted price.
The five tiers: Free ($0, 1,000 credits, Sonnet only) for trying the Forge. Studio ($49/mo, 10,000 credits) for solo founders. Team ($199/mo, 50,000 credits, 10% off top-ups) — most popular. Pro ($499/mo, 200,000 credits, 20% off top-ups, priority queue) for power users. Enterprise for SSO, dedicated capacity, and a CSM. Every paid tier can use Opus 4.7 — the credit cost (~5× Sonnet) is the natural gate, no artificial lockouts. Tiers cap how often you can run and how much your top-ups cost.
Why credit packs. Builds vary wildly in cost. A landing page is cheap. A multi- persona platform with auth, integrations, and a polished demo is not. Capping every plan at “X builds per month” forces us to either subsidize the heavy ones (which kills margin) or under-deliver on the light ones (which kills trust). Credit packs let everyone pay for what they actually use — same shape Anthropic charges us in.
Margin.We mark up upstream API cost ~8× — the SaaS-on-API industry-standard band. That covers the council orchestration overhead, retries, the polish-audit pass, ElevenLabs voice on voice-bearing builds, hosting, support, and Stripe fees, and leaves us ~75% net gross margin at full utilization. Lovable, ChatGPT and Claude Pro all sit in the same band. Edits run lighter (6× margin) because they reuse a build's cached system block.
We're holding everything FREE during launchso we can learn what people actually build. Once we flip Stripe on, the prices above are what they'll be.
