Claude tokens resold at up to 93% off — what you actually get
A relay market moves LLM access at 90%-plus discounts. The price of those savings: swapped-in models, harvested prompts, and accounts built to get banned.
Copy markdown
The discounts that pull builders in
Resold Claude tokens run 70-93% below list, Codex access as low as 1-5% of API cost, and median discounts hit 94-98% across the relay chain. That's the bait now sitting in Discord and Telegram DMs.
You may not get the model you paid for
Independent tests found 'Gemini-2.5' served through gray-market proxies scored 37% on medical questions versus 83.82% on Google's real API — silently swapped for a cheaper or distilled stand-in. You could burn a day debugging a model you never actually called.
Your prompts become someone's training set
The proxies sit in the middle and log every request, then sell the traces for distillation. Anything you send — source code, API keys, customer data — is the actual product being resold.
The accounts are built to be banned
The supply runs on stolen cards and fake IDs: Anthropic flagged roughly 25,000 fraudulent accounts and 16M+ extracted exchanges in a single sweep. Access bought this way can vanish mid-project when the pool gets nuked.
How the relay actually works
Four layers stacked: card merchants fund the fraud, account pools bundle Claude Max subscriptions, One-API panels proxy the traffic, and buyers plus distiller labs sit at the end. Recognizing that stack is how you spot a reseller's pitch.
The legit way to cut your bill
Marketplaces now broker genuinely unused OpenAI, Anthropic, and Gemini credits, and the labs answered the fraud with ID-plus-biometric verification back in July. Cheaper access without the ban risk exists — the 90%-off DM isn't it.