Pricing

Vibe-code with millions of tokens.

Every plan unlocks every Claude model. Pay monthly, get hundreds of millions of tokens. Failed calls cost nothing. Sign in with Telegram, pay in crypto, ship today.

Pro

Popular
For shipping side projects
$9.99/ month
500M tokens each cycle
Get Pro
No commitment · Cancel anytime
  • Every Claude model — Haiku, Sonnet, Opus
  • 500M tokens per month — vibe-code freely
  • 10 concurrent requests
  • Failed calls are free, always
  • Drop-in Anthropic + OpenAI SDKs

Ultra

For builders going full-time
$19.99/ month
1.5B tokens each cycle
Get Ultra
No commitment · Cancel anytime
  • 1.5B tokens — 3× Pro
  • 15 concurrent requests
  • Priority routing queue
  • Higher per-call token caps
  • Early access to new models

Max

For teams and agent fleets
$39.99/ month
5B tokens each cycle
Get Max
No commitment · Cancel anytime
  • 5B tokens — 10× Pro
  • 25 concurrent requests
  • Priority queue + dedicated capacity
  • Custom rate-limit policies
  • Direct support channel

Models & rates.

Token rates are computed from how expensive each model is upstream. The same call costs the same on every plan.

Save up to 98% vs the official per-token rate.
Every plan unlocks every model. Pro gives you 500M plan tokens for $9.99 — same calls, fraction of the price.

Haiku

Fast, light, cheapest
ModelPlan tokens / 1M API inPlan tokens / 1M API out
Claude Haiku 4.5
claude-haiku-4-5-20251001
1M5M

Sonnet

Best price / quality
ModelPlan tokens / 1M API inPlan tokens / 1M API out
Claude Sonnet 4.6
claude-sonnet-4-6
3M15M
Claude Sonnet 4.5
claude-sonnet-4-5-20250929
3M15M

Opus

Most capable
ModelPlan tokens / 1M API inPlan tokens / 1M API out
Claude Opus 4.7
claude-opus-4-7
15M75M
Claude Opus 4.6
claude-opus-4-6
15M75M
Claude Opus 4.5
claude-opus-4-5-20251101
15M75M
How rates work

Pro ships with 500M plan tokens / month. The numbers above show how many of those a million API tokens uses. So 1M Sonnet output tokens uses 15M of your budget — about 3% per million. Opus output is the priciest at 75M per million — but Pro still gives you ~6.6M Opus output tokens, plenty for serious vibe coding.

Cached input billed at 10% of input. Cache writes at 125% of input. Failed calls cost 0 tokens.

Frequently asked questions.

What's a 'plan token'?
Each plan ships with a monthly pool of plan tokens that you spend on any Claude model. Rates are measured in Haiku-input-equivalent tokens, so 1 plan token covers exactly 1 Haiku input API token. More expensive models consume the pool proportionally faster (Sonnet input = 3×, Opus output = 75×, etc.). It's the same idea as the official pricing — just rebranded so the headline number on each plan looks like a sensible "tokens / month".
What happens to unused tokens at the end of the month?
Tokens expire 30 days after they're issued. They don't roll over — that's how we keep flat-rate pricing this cheap.
What if a request fails?
Failed calls cost zero tokens. We only debit on a successful 2xx response (or a stream that completes cleanly with message_stop). Upstream errors, timeouts, or stream aborts trigger a 100% automatic refund.
Which Claude models can I use?
Six Claude models — Haiku, Sonnet, and Opus across their dated revisions — all verified live with our upstream account. Every paid plan unlocks every model; pick per call.
Is this compatible with the official Anthropic SDK?
Yes — TypeGPT speaks the same /v1/messages protocol. Swap base_url to ours, use Authorization: Bearer sk-typegpt-…, done. We also expose an OpenAI-compatible /v1/chat/completions adapter for tools that expect that shape.
How do I pay? Can I get a refund?
Payments go through OxaPay — BTC, USDT (TRC20/ERC20/BEP20), LTC, TRX, USDC, DOGE, TON, BNB, ETH. No card required, works worldwide. We honor refund requests within 24 hours of purchase if you've used less than 10% of the cycle's tokens.
Is TypeGPT affiliated with Anthropic?
No. TypeGPT is an independent API gateway. We forward requests to Claude's API and bill you in plan tokens. "Claude" is a trademark of Anthropic, PBC.
How is my data handled?
Zero retention on both sides. We never store your prompts or responses — and Anthropic doesn't either, because TypeGPT runs on a Zero Data Retention (ZDR) enterprise workspace upstream. We keep only call metadata (model, token counts, HTTP status, latency) for your usage logs and billing. Read the full TypeGPT privacy policy.
More questions? Message @TypeGPTAPIBOT or email [email protected].