Token Cost to Build an App Like Netflix with Claude Sonnet 4.6 (2026)
Last updated: 12 May 2026Model: Claude Sonnet 4.6Data source: MyAppTemplates.com analysis of 2026 public SOW benchmarks and shipped-app case studies.
Executive Summary
A Netflix-style streaming app is a license-gated build: the software scope is well-trodden — auth, catalogue, player, subscriptions — but the real cost lives in content rights, DRM, and CDN egress. This page isolates the software-only build, phase by phase, and measures it in Claude Sonnet 4.6 token spend against the $199 MyAppTemplates boilerplate.
Claude Sonnet 4.6 is the right default model for roughly 80% of this build — it's fast, it's cheap ($3 / $15 per Mtok input/output as of 2026), and it handles Drizzle schemas, Hono routes, and Expo screens cleanly. The 20% that genuinely benefits from escalating to Opus 4.5 is the architectural set: DRM token-exchange flow, adaptive-bitrate playback state machine, and the entitlement model that gates which titles a user can actually stream. Everything else is Sonnet.
Total software-only AI spend across all six phases lands at $275–$350 over 2–3 weeks of focused work. Mid-market agency quotes for the equivalent software scope (excluding licensing, CDN contracts, and content deals) typically land at $150k–$250k. Both numbers are accurate; they are not the same buyer.
Phase-by-phase token math
Netflix clone build phases with Claude Sonnet 4.6
Software scope only — content licensing, DRM contracts, and CDN egress excluded.
Every DIY build starts with the same flat boilerplate fee:$199 one-time — column below shows marginal Claude Code API spend on top
Deploy & CIWorkers deploy, EAS builds, Sentry, GitHub Actions
Sonnet 4.6 only
$10k–$20k
$10
Pre-wired
0.5 day
1. Where Sonnet 4.6 is the right tool
Sonnet 4.6's strengths are the bulk of a Netflix clone: typed schemas, REST routes, Expo screens, and CRUD-ish flows where the architecture is already decided. At $3 / $15 per Mtok, you can iterate on UI states and route handlers without watching the meter. The boilerplate's modular routes pattern (routes/example-routes.ts) keeps each feature scoped, so Sonnet doesn't need to load the whole codebase into context to add a watchlist endpoint.
Spotlight Phase
Catalogue schema in 1 day, ~$25 in tokens
ScopeTitles, episodes, seasons, genres, cast, watchlists, watch-history, ratings — six Drizzle tables with foreign keys and indexes.
Token mix~5M input / ~500k output across 3–4 Claude Code sessions.
Cost$25Sonnet 4.6 only
Why Sonnet winsSchema design is pattern-matching against the existing db/schema.ts conventions — no novel architectural decisions, so Opus is overkill.
Spotlight Phase
Home + title detail + search UI in 4 days, ~$85
ScopeHorizontal-row home screen, title detail with episode list, full-text search, profile picker, downloads list.
Token mix~20M input / ~1.8M output across the week — most of it re-reading existing tab and component files.
Cost$85Sonnet 4.6 only
Why Sonnet winsExpo Router + tab navigation are already scaffolded. Sonnet extends the existing patterns instead of inventing new ones.
2. Where to escalate to Opus 4.5
Roughly one phase out of seven genuinely benefits from Opus: the player and DRM glue. The reason is not implementation difficulty — it's the number of moving pieces that must agree. License token exchange, adaptive-bitrate state, offline-download key storage, resume-marker sync, and the entitlement check all interact. Opus is worth its $15 / $75 per Mtok for the architectural pass, then Sonnet implements the resulting plan.
Spotlight Phase
Player & DRM: ~$70 with a hybrid Opus+Sonnet workflow
PatternOne Opus session to design the token-exchange + ABR + offline-key state machine (~$25). Sonnet then writes the React Native bridge, Workers endpoint, and tests against that design (~$45).
Boilerplate roleThe rate-limited endpoints and JWT session handling are already in place — you wire the DRM license-server call against them. Cloudflare Workers is the runtime; you add the license-proxy route as a feature module.
License caveatWidevine, FairPlay, and PlayReady licences themselves are commercial agreements with content owners. The software glue is what's $70; the licences are separate.
3. What the $199 + $275–$350 number does not cover
The software-only framing is honest only if you read what it excludes. Streaming is a content-rights business with a thin software layer on top — the build numbers on this page are the thin layer.
Spotlight Exclusion
Content licensing & CDN egress
Content rightsStudio licensing deals start at six figures for a small catalogue and scale into the millions for anything resembling a mainstream library. None of this is in the software build cost.
CDN egressVideo bytes are the dominant infra line. At Netflix-style watch hours, egress runs $0.005–$0.02 per GB across Cloudflare R2 / CloudFront / Bunny, dwarfing Workers compute.
TranscodingMux, AWS MediaConvert, or Bitmovin handle ABR ladders and DRM packaging — usage-priced, not in the boilerplate. The billing abstraction can accept usage adapters; you wire the metering.
How to run a Netflix-clone build with Sonnet 4.6 as default
A repeatable workflow that keeps Sonnet on the hot path and uses Opus only where it pays for itself.
1
1. Clone the boilerplate and ship a build the same day
Auth, billing adapters, Workers runtime, Drizzle schema, Sentry, and CI are already wired. Run /db-migrate, deploy to Workers, build with EAS. You now have a signed-in app with a paywall before you've written a feature.
2
2. Use Sonnet 4.6 for schema, routes, and UI
Open Claude Code with the @backend-dev subagent for schema/routes and @mobile-dev for screens. Stay on Sonnet. Use /new-feature to scaffold catalogue, search, watchlist, and history as feature modules.
3
3. Escalate to Opus only for the player architecture
Spend one Opus session designing the DRM token exchange, ABR state machine, and offline-key flow. Save the plan as a markdown file in the repo. Switch back to Sonnet to implement against that plan.
4
4. Wire subscriptions against the existing billing adapter
The Stripe and RevenueCat adapters are already in place. Define your plan tiers, point the paywall screen at them, and gate streaming with entitlement checks. Half a day with Sonnet.
5
5. Pick your CDN and transcoder before scaling traffic
Mux + Cloudflare Stream or Bunny CDN are the pragmatic choices for solo builders. Wire usage metering into the billing abstraction so per-user egress is observable from day one.
Frequently Asked Questions
Why default to Sonnet 4.6 instead of Opus for the whole build?
Cost-per-token, mostly. Sonnet runs at $3 / $15 per Mtok versus Opus at $15 / $75. For Drizzle schemas, Hono routes, and Expo screens — which is 80% of a Netflix clone — Sonnet's output quality is indistinguishable. Opus earns its premium on multi-system architectural decisions, not on writing a watchlist endpoint.
Is $275–$350 in tokens realistic for a 2–3 week build?
Yes, for a focused solo build using Claude Code agentically against the boilerplate. Most token spend is re-reading existing files for context, not generating novel code. The boilerplate's modular routes pattern keeps the relevant context small per session, which is why the number is this low.
Does the boilerplate include video playback or DRM?
No. The boilerplate gives you auth, billing, Workers runtime, Drizzle schema, and a working mobile shell. Video playback, DRM licence exchange, and CDN integration are features you build on top — Sonnet 4.6 plus a single Opus architectural pass handles the glue cleanly.
What's actually pre-wired for subscriptions?
The Stripe and RevenueCat adapters, a billing abstraction layer, a subscription schema, a paywall screen, and a mock provider for local development. You define your plan IDs and entitlement rules; you don't write the integration.
Why is the agency quote range $150k–$250k?
That's the typical mid-market band for the software scope of a streaming app — auth, catalogue, player glue, subscriptions, admin tooling — before licensing, DRM contracts, CDN, and content costs. Mid-market agencies are pricing delivery, project management, QA, and warranty, which is real work.
Can I run this entirely on Sonnet 4.6 and skip Opus?
You can, and many builders do. The DRM and player phase will take longer because Sonnet will iterate more on the state machine. Expect 5–6 days instead of 3–4, and add maybe $30 in extra tokens. Skipping Opus is a defensible choice for solo builders watching every dollar.
Sonnet 4.6 plus $199 of scaffolding covers the software scope. Licensing is a different conversation.
For the part of a Netflix clone you can actually code, the honest number is $199 boilerplate + $275–$350 in Sonnet 4.6 tokens over 2–3 weeks, with one Opus session for the player architecture. That gets you a deployable streaming app shell with auth, catalogue, subscriptions, and a working player. The content rights, DRM contracts, and CDN are still ahead of you — and they always were, regardless of who built the software.