Token Cost to Build an App Like Amazon with Claude Sonnet 4.6 (2026)
Last updated: 9 May 2026Model: Claude Sonnet 4.6Data source: MyAppTemplates.com analysis of 2026 public SOW benchmarks and shipped-app case studies
Executive Summary
An Amazon-style commerce app is a wide-scope build — catalogue, search, cart, checkout, payments, order state, returns, reviews, addresses, and a seller surface if you go multi-vendor. This page measures the actual Claude Sonnet 4.6 token cost per build phase, on top of the $199 MyAppTemplates boilerplate, for a solo founder shipping a credible v1 in roughly 9–14 days.
Sonnet 4.6 is the right default for ~80% of this build: schema work, route handlers, UI screens, list rendering, form wiring, billing-adapter glue, and deploy scripts. Escalate to Opus 4.5 only for architecture decisions — checkout state machine, inventory consistency, search ranking, payments reconciliation. The split is roughly 90% Sonnet tokens, 10% Opus tokens by volume, which is what keeps a single-founder build under $300 in API spend.
Mid-market agency quotes for an Amazon-style consumer app with payments and a seller dashboard typically land at $70k–$150k — that pays for delivery, QA, project management, and warranty. The DIY route is a different buyer profile: a hands-on founder who wants to own the codebase and iterate against real users. The numbers below are for that buyer.
Token math
Per-phase Claude Sonnet 4.6 spend for an Amazon-style build
Phases run roughly in order. Token counts are agentic-loop totals, not single completions.
Every DIY build starts with the same flat boilerplate fee:$199 one-time — column below shows marginal Claude Code API spend on top
Sonnet 4.6 is fast and cheap enough to run as the default coding loop. For an Amazon-style app, the heavy lifting is repetitive surface work — schema rows, route handlers, list screens, form wiring — which is exactly the work Sonnet handles at 99% acceptance with the boilerplate's existing patterns to copy from.
Spotlight phase
Catalogue & product detail screens
Sonnet token cost$34across ~1.5 days of agentic work
Pattern sourceBoilerplate's example-routes.ts and tab navigation give Sonnet a working CRUD reference to extend.
Why Sonnet, not OpusList → detail → list → detail is mechanical. No architectural decisions, just consistent application of an existing pattern.
Spotlight phase
Mobile UI for the buyer surface
Sonnet token cost$567 screens, ~2.5 days
What you getHome, search results, PDP, cart, checkout, order history, profile — all wired against the boilerplate theme system and tab nav.
Tooling that helpsThe @mobile-dev subagent and /new-feature slash command keep file layout consistent across screens.
2. Where to escalate to Opus 4.5
About 10% of token volume should be Opus. The rule of thumb: any decision that, if wrong, costs you a refactor across multiple files. On a commerce app, that's a short and predictable list.
Escalate
Checkout state machine
Why OpusCart → address → payment → confirm has branching (guest checkout, saved address, applied promo, retry on payment failure). Get the state model wrong once and every screen leaks logic.
Token mix70% Sonnet implementing / 30% Opus designing the state diagram and reviewing edge cases
Escalate
Inventory & order consistency
Why OpusTwo buyers can't both buy the last unit. The decision about where you enforce that — at cart-add, at checkout, at payment-success — has D1 schema and Workers concurrency implications.
What Opus producesA short ADR-style note in your repo, then Sonnet implements it.
Escalate
Payments reconciliation
Why OpusWebhooks arrive out of order. Refunds, chargebacks, and partial captures need a reconciliation strategy that survives reboot. This is one Opus session, then Sonnet implements the handlers.
Token mix85% Sonnet / 15% Opus
3. What the boilerplate removes from this number
The token totals above assume you're starting from the MyAppTemplates boilerplate. If you weren't, the first $500–$1,500 of Sonnet spend would go on rebuilding work that's already done — auth screens, JWT middleware, billing abstraction, Drizzle schema scaffold, Workers config, CI, Sentry, theme system. None of that is Amazon-specific.
BillingAdapter pattern with Stripe and RevenueCat adapters. For Amazon-style one-time payments, you extend the Stripe adapter rather than building from scratch.
BackendCloudflare Workers + D1 + Drizzle + Wrangler config. CI on GitHub Actions. Sentry scaffolded.
AI toolingAGENTS.md, CLAUDE.md, @backend-dev and @mobile-dev subagents, slash commands. Claude Code is productive in the repo from minute one.
How to run the build
A repeatable order of operations for a solo founder using Claude Code with Sonnet 4.6 as default and Opus 4.5 on call.
1
Day 1 — Schema and seed
Open Opus once to lock the schema (products, variants, orders, sellers, reviews, addresses). Switch to Sonnet to write Drizzle migrations and a seed script. Run /db-migrate. Cost: ~$22.
2
Days 2–3 — Catalogue and search
Sonnet builds list, filter, facet, pagination, and PDP routes plus the matching mobile screens. Use /new-feature catalogue. Cost: ~$90.
3
Days 4–5 — Cart, checkout, payments
Open Opus to design the checkout state machine and payments reconciliation. Sonnet implements both ends and extends the Stripe adapter. Cost: ~$86.
4
Days 6–7 — Order lifecycle, reviews, profile
Pure Sonnet work — order status transitions, reviews schema and flows, address book, profile screen. Cost: ~$50.
5
Days 8–10 — Seller surface and deploy
Optional seller dashboard if you're going multi-vendor. Wrangler deploy, Sentry verification, smoke tests through CI. Cost: ~$54.
Frequently Asked Questions
Is $300 of Sonnet 4.6 spend really enough to ship an Amazon-style app?
For a credible v1 with a buyer surface, payments, orders, and reviews — yes, on top of the $199 boilerplate. The number assumes you accept Sonnet's first or second pass on most files and reserve Opus for state-machine and consistency decisions. If you re-prompt aggressively or ignore the boilerplate's existing patterns, expect 2–3x.
Why not just use Opus 4.5 for everything?
You'd pay roughly 5x more for marginal quality on routine surface work. Opus is worth it where the cost of a wrong decision is a refactor — checkout state, inventory consistency, payments reconciliation. For list screens and form wiring, Sonnet matches Opus's accepted output rate on this codebase.
Does the boilerplate include marketplace features like multi-vendor payouts?
No. The billing abstraction accepts Stripe Connect as an adapter, but the Connect integration itself is your work. Plan an extra 1–2 days and ~$60 of Sonnet spend if you're going multi-vendor with seller payouts.
What about search ranking and recommendations?
Basic SQL-driven search and filters are in scope of the numbers above. Algolia, vector search, or recommendation models are external integrations — budget separately. For a v1, postpone them; Amazon shipped without recommendations for years.
How does the agency benchmark of $70k–$150k break down?
Mid-market agency SOWs for this scope typically allocate 40–50% to engineering, 15–20% to design, 10–15% to QA, and the rest to project management, account management, and warranty. The DIY route trades that delivery package for your own time and a working codebase you own end-to-end.
What's the realistic timeline including testing and App Store review?
9–14 days of build, then 5–10 days for App Store and Play Store review on first submission. Plan for 3 weeks calendar time from clean repo to live downloads.
Sonnet does the work. Opus does the architecture. The boilerplate does the first week.
An Amazon-style v1 with Claude Sonnet 4.6 lands at roughly $300 of API spend over 9–14 days, on top of the one-time $199 boilerplate. Opus 4.5 shows up for three or four short sessions and earns its premium on the decisions that would otherwise cost you a refactor.