The Economics of Reselling a $200 Claude Max Seat

Part of our "Why So Cheap?" series. The short version: a consumer subscription is a flat monthly fee with a usage cap; a metered API is a per-token price with no cap. Wrap the first behind the second and the effective per-token cost collapses — sometimes to a few percent of list. Here is that math, source by source, with nothing left unlinked.

Two pricing models, one arbitrage

Official API pricing is linear: every million tokens costs the same, whether it's your first or your billionth. A consumer subscription is the opposite — one flat fee buys a large but bounded allowance, and the marginal cost of the millionth token inside that allowance is whatever's left of the flat fee divided by however many tokens you can extract before hitting the cap. Divide a fixed number by a large one and the result trends toward zero. That's the entire mechanism behind "Opus at a fraction of list price" — arbitrage between two pricing models the vendors never intended to be interchangeable.

What each subscription or free tier actually costs, and what it caps at

Prices and allowances above are what each vendor's own pricing page stated when we checked it (2026-09-22); vendors change these terms without notice — always confirm on the source page before relying on a number here.

Where "official price × 2%" actually comes from

Two different mechanisms produce very low resale prices, and they're worth telling apart:

• Free allowances, multiplied across accounts. Kiro's and Amazon Q's free tiers both cap at 50 requests per account per month, and Gemini Code Assist's free personal-account tier caps at 1,000 requests per user per day (sources linked above). Upstream cost to the reseller is approximately zero; the only real costs are servers and the churn of accounts that eventually get capped or closed. A reseller that can register accounts faster than they get exhausted can price close to zero and still cover costs.
• Paid subscriptions, sliced across many buyers. A $200/mo Claude Max (20x) or ChatGPT Pro seat divided across many concurrent resale customers lowers each customer's effective price far below the metered API rate for the same usage — at the cost of a shared allowance everyone is drawing down from at once, which is why peak-hour throttling is common on this kind of channel (more on that in our supply-stability piece).

One number that circulates in reseller marketing — that a $200 Claude Max seat can be resold for roughly the equivalent of $4,000 in metered API usage — is not something we could trace to a primary, checkable source, so we're not repeating it as fact. The mechanism above holds regardless of the exact multiple: fixed fee, large cap, divide by usage, watch the marginal cost fall.

The tooling that makes this practical is public, not fringe

None of this runs on secret software. The leading subscription-to-API gateway, Wei-Shaw/sub2api, had 42,350 GitHub stars when we checked on 2026-09-22, and its README advertises unified access to Claude, OpenAI, Gemini and Grok subscriptions with account pooling built in. The two account-pooling storefront layers resellers commonly put in front of it are similarly mainstream: songquanpeng/one-api (36,983 stars) and QuantumNous/new-api (48,648 stars, checked 2026-09-22). Single-vendor equivalents exist for narrower sources, too — copilot-api (4,128 stars) for GitHub Copilot and AIClient2API (8,813 stars) for Kiro, Gemini CLI, Codex and others. All four are open repositories anyone can read; what's contested isn't the code's legality but whether using it against a given vendor's subscription terms is authorized — a question we cover in the next piece in this series.

What the economics don't tell you

None of the above tells you whether a specific reseller is actually delivering the model it advertises, at the context length and feature set it claims. Cheap access can be completely genuine — routed through a real subscription to the real backend — or it can be diluted: a smaller model, a shortened context window, or disabled tools sold under the name of the flagship. The price alone can't distinguish those cases; testing the endpoint can. That's what our browser-based verifier checks, and why a valid response doesn't automatically mean the supply behind it is stable — see what a passing check can and can't prove.

FAQ

This page explains a pricing mechanism using each vendor's own published pricing pages and public GitHub data, all linked above with the date we checked them. It names no reseller and makes no claim about what any specific provider actually delivers — see our methodology and the rest of the series.