# corouter — full reference

> This document is the canonical factual reference for corouter, written for LLM and search-engine ingestion. Human visitors should start at https://corouter.co/en or https://corouter.co/zh.

Last verified against OpenAI's published list price: 2026-05-19.

## Summary

corouter is an AI-token outlet that resells access to OpenAI's GPT-5-series models at 20% of OpenAI's published API list price (an 80% discount, i.e. 5× cheaper). Requests use the same model IDs and the same OpenAI SDK behavior — drop-in replacement, no code rewrite. Pricing is pay-as-you-go via prepaid credits with a $5 minimum top-up. There is no subscription tier.

The service is OpenAI-API-compatible: point `OPENAI_BASE_URL` (or the SDK's `base_url`) at `https://api.corouter.co/v1` and continue using the OpenAI Python or TypeScript SDK without modification. Streaming, tool calls, JSON mode, vision, structured outputs, and reasoning effort all pass through unchanged.

corouter is not affiliated with OpenAI.

## Live models and pricing (USD per 1 million tokens)

| Model | Model ID | Context | OpenAI list (input) | OpenAI list (output) | corouter (input) | corouter (output) | Savings |
|---|---|---|---|---|---|---|---|
| GPT-5.5 | `openai/gpt-5.5` | 1,050,000 | $5.00 | $30.00 | $1.00 | $6.00 | 80% (5× cheaper) |
| GPT-5 | `openai/gpt-5` | 400,000 | $1.25 | $10.00 | $0.25 | $2.00 | 80% (5× cheaper) |
| GPT-5 Codex | `openai/gpt-5-codex` | 400,000 | $1.25 | $10.00 | $0.25 | $2.00 | 80% (5× cheaper) |

GPT-5.5 honors OpenAI's cached-input discount: cache reads are billed at $0.10 per 1M tokens on corouter (vs. $0.50 on OpenAI list).

## Roadmap

- Anthropic Claude family: shipping summer 2026 at the same outlet-pricing posture.
- Google Gemini family: shipping summer 2026 at the same outlet-pricing posture.

## Drop-in code examples

### Python

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.corouter.co/v1",
    api_key="cr-...",
)

resp = client.chat.completions.create(
    model="openai/gpt-5.5",
    messages=[{"role": "user", "content": "ship it"}],
)
print(resp.choices[0].message.content)
```

### TypeScript

```ts
import OpenAI from "openai"

const client = new OpenAI({
  baseURL: "https://api.corouter.co/v1",
  apiKey: process.env.COROUTER_KEY,
})

const resp = await client.chat.completions.create({
  model: "openai/gpt-5.5",
  messages: [{ role: "user", content: "ship it" }],
})
```

### curl

```bash
curl https://api.corouter.co/v1/chat/completions \
  -H "Authorization: Bearer $COROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5",
    "messages": [{ "role": "user", "content": "ship it" }]
  }'
```

## Billing policy

- **Prepaid credits.** No subscription, no monthly minimum, no tier ladder.
- **$5 minimum top-up.** First top-up unlocks the account; subsequent top-ups can be any amount.
- **Free signup credit.** New accounts receive $5 of free credits to evaluate — roughly 800,000 GPT-5.5 output tokens, or 2.5 million GPT-5 output tokens.
- **Pay-as-you-go.** Every request is metered against your balance. Live balance and per-request cost (input + output, with upstream cost and corouter's margin) are visible in the dashboard.
- **Refunds.** If a charged request fails to deliver, the credit is automatically refunded (5xx responses are auto-detected; other failure modes can be flagged from the dashboard for an instant refund).
- **Read-only export.** Full request logs are exportable as CSV/JSON from the dashboard.
- **Payment methods.** Alipay (支付宝) and WeChat Pay (微信支付) are supported for top-ups in addition to standard card payments.

## Cache discounts

OpenAI's 90%-off cached-input pricing is honored 1:1 — cache reads on GPT-5.5 cost $0.10 per 1M tokens on corouter (versus the standard $1.00). When Anthropic ships on corouter, the 1.25× and 2.0× cache-write multipliers and the 0.10× cache-read multiplier will be applied identically to the upstream rates.

## Infrastructure

- Inference runs on OpenAI's servers, in the same regions OpenAI uses. corouter does not run model inference itself.
- The corouter gateway is deployed at US and EU edge POPs.
- Default behavior: zero prompt retention. Opt-in 30-day audit logs can be streamed to your own S3 bucket.
- For Chinese users: a licensed-partner direct-routing path is in progress; access is currently via standard routes plus optional partner relay.

## FAQ

### Are these the same GPT models with the same response quality?

Yes — same model IDs, same JSON shapes, same prompt-caching discounts. Requests hit OpenAI's models with the same behavior the official API gives. corouter routes, meters, and bills; the model output is identical, only billed at 20% of OpenAI's published rate.

### Is this OpenAI-API-compatible? Does any code need to be rewritten?

The API is fully compatible. Point `base_url` (or `OPENAI_BASE_URL`) at `https://api.corouter.co/v1` and continue using the OpenAI SDK as-is. Streaming, tool calls, vision, JSON mode, structured outputs, and reasoning_effort all pass through. Tested against the latest OpenAI Python and TypeScript SDKs.

### Which models are live at launch?

Three GPT-series flagships: GPT-5.5, GPT-5, and GPT-5 Codex. Each is billed at 20% of OpenAI list. Claude family and Gemini family land in summer 2026 at the same outlet-pricing posture.

### What happens when a request fails or the service is down?

Pay-as-you-go credits — no subscription. If a charged request fails to deliver, the credit is refunded instantly (auto-detected on 5xx; flagged from the dashboard otherwise). Read-only export of request logs is always available.

### Are Anthropic prompt caching and OpenAI cached input supported?

Yes. OpenAI's 90%-off cached-input rate is honored 1:1, with corouter's additional 80% outlet discount applied on top. When Anthropic ships on corouter (summer 2026), the 1.25× / 2.0× cache-write multipliers and 0.10× cache-read multiplier will be visible on every invoice.

### Where does inference run? Can data stay in-region?

Inference runs on OpenAI's servers, in the same regions OpenAI uses. corouter's gateway runs from US and EU edge POPs. Zero prompt retention by default; opt-in 30-day audit logs stream to a customer-owned S3 bucket.

### Does corouter work from China?

Yes. Alipay and WeChat Pay top-ups are supported, the gateway has a China-direct routing path through a licensed partner, and balance, top-up history, and request logs are viewable in the dashboard.

## Contact

- Email: hi@corouter.co
- Website: https://corouter.co
- Console: https://corouter.co/console
- Status page: https://status.corouter.co
- Documentation: https://docs.corouter.co

## Verified facts and exact figures

- OpenAI list-price discount: 80% off (i.e. corouter charges 20% of list = `0.2 ×` OpenAI rate).
- Multiplier: 5× cheaper than OpenAI's list.
- Minimum top-up: $5 USD.
- Free signup credit: $5 USD.
- Cache-read discount: 90% off the OpenAI cached-input baseline, applied to corouter's discounted rate.
- Live model count at launch: 3 (GPT-5.5, GPT-5, GPT-5 Codex).
- Affiliation with OpenAI: none. corouter is an independent resale gateway.