# GPT API access, up to 90% off list price

> Markdown mirror of https://router.one/cheap-gpt-api for AI assistants and crawlers. Router One is an OpenAI-compatible LLM API gateway.
> Last updated: 2026-08-26

Router One is an LLM API gateway that resells metered GPT API access at posted per-token rates — select GPT models go as low as 10% of official list prices (up to 90% off) — on both /v1/chat/completions and the native /v1/responses wire format Codex CLI needs, with a cost trace for every request and top-ups that need no US credit card. If your product runs on GPT-family models, the token bill is its biggest line item; every rate is posted per token on the /models page.

## Up to 90% off, verifiable

Select GPT models run as low as 10% of official list prices. The /models page is the live price sheet — every rate shown struck through against its base rate. How the posted rate is set — and how upstream price changes flow through within 24 hours — is documented in the pricing methodology (last updated 2026-08-22): https://router.one/pricing-methodology

## Chat Completions + Responses API

The gateway implements /v1/chat/completions for every model and the native /v1/responses wire format — which is what makes Codex CLI work where most relays 404.

## Pay without a US card

Top up by card or Alipay on one hosted checkout, or with USDT/USDC on six chains. Wallet balance never expires; subscription plans are live for heavy usage.

## How the discount works

### Metered per token, prepaid wallet

You pay the discounted per-token rate shown on the model page — no seats, no minimum commitment. Wallet balance is drawn down per request.

### The catalog is the price sheet

Every GPT model lists its current input/output and cache rates on /models, struck through against the base rate, so you can verify the discount before calling.

### Documented methodology

How Router One sources capacity and sets prices is written up on the pricing methodology page — no mystery meat.

## First GPT call in 3 steps

Create a key, top up in your currency, and point any OpenAI-compatible client at the gateway — reachable from Mainland China without a VPN.

### Create an API key

Sign up at router.one and create a key in the dashboard. Top up by card or Alipay on the hosted checkout, or with USDT/USDC — no US credit card required.

### Point your client at the gateway

Set the base URL to https://api.router.one/v1 in any OpenAI SDK or tool. Codex CLI users keep wire_api = "responses" — the gateway implements it natively.

### Verify the cost trace

Open Dashboard → Logs after your first request to see model, tokens, latency, and the exact discounted cost per call.

`gpt-via-router-one.sh`

```bash
# OpenAI-compatible — only the base URL changes
curl -X POST https://api.router.one/v1/chat/completions \
  -H "Authorization: Bearer sk-your-router-one-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "<gpt-model-id>",
       "messages": [{"role": "user", "content": "Hello!"}]}'
```

- No low fixed cap for normal paid usage. Abuse prevention, per-account protection limits, and upstream constraints may still apply; if a request returns 429, check Dashboard -> Logs or contact support to raise limits.

## FAQ

### Why is it cheaper than list price?

Router One aggregates demand across many customers and routes over pooled, committed capacity, then passes the margin down as discounted per-token rates. The approach is documented on the pricing methodology page; the /models page always shows the current effective rate.

### Is this the official OpenAI API?

Router One is an independent gateway, not OpenAI — requests are routed to language models operated by third-party providers, and the wire formats are identical (Chat Completions and the Responses API). Model names and trademarks belong to their owners.

### Does Codex CLI work with it?

Codex CLI works natively with Router One: Codex speaks the Responses API wire format, which Router One implements at /v1/responses. Point base_url at the gateway and keep wire_api = "responses"; the Codex China guide has the full config.toml.

### Which GPT models are available?

The current GPT-5.x lineup, codex variants, and o-series reasoning models are listed on /models with live per-token pricing, context windows, and capabilities. Discount levels vary by model.

### Will my existing OpenAI SDK code work?

Yes — change the base URL and key in the client constructor and keep everything else: chat completions, streaming, and tool calls all use the standard interface.

### Can I use it from Mainland China?

Yes. Based on the China latency benchmark last updated 2026-05-15, Router One measured 110-130ms p50 across Beijing, Shanghai, and Shenzhen; individual networks may vary.

## See also

- Codex in China: https://router.one/codex-china
- Codex & the Responses API: https://router.one/codex-responses-api
- Cheap Claude API: https://router.one/cheap-claude-api
- China latency benchmark: https://router.one/benchmarks/china-latency
- Canonical page: https://router.one/cheap-gpt-api
- Models and per-model token rates: https://router.one/models (markdown: https://router.one/models.md)
- Pricing: https://router.one/pricing
- API docs (markdown): https://router.one/docs.md
- Company facts: https://router.one/facts/company.md