Claude / GPT API relay: low cost, stable, genuine
OrcAI puts the Claude and GPT model families behind one Anthropic / OpenAI compatible endpoint. Pay as you go with ¥1 = $1 of credit, top up from ¥50, credits that never expire; Claude Code, Codex CLI and Cursor connect by changing one BASE_URL.
Why route Claude / GPT through OrcAI
Four claims, each verifiable in the console, the status page and your own usage logs.
¥1 buys $1 of credit
Every ¥1 you top up becomes $1 of credit, billed at list price × group multiplier — far cheaper than buying USD at the exchange rate. Current group multipliers are shown in the console, no hidden markup.
2.14M requests in 7 days
Multi-account pools fail over automatically; every model is probed every 90 seconds with latency and availability published on the status page. Availability target 99.9%.
Genuine models, never swapped
Ask for Opus 5 and you get Opus 5: the model field and usage come back untouched, no downgrade, no swap, no context truncation. Zero swaps across 1M+ logged responses in 7 days; three self-checks are documented below.
Current Claude and GPT lineups
Claude Opus 5 / Sonnet 5 / Opus 4.8 / Sonnet 4.6 / Haiku 4.5, GPT-5.6 / 5.5 / 5.2; IDs match the vendors', new models arrive as they ship.
Supported models and groups
Two public groups, IDs identical to the vendors'. Probe status refreshes every 5 minutes; the full list is on the models page.
| Group | Model ID | Vendor | List price | Probe status |
|---|---|---|---|---|
| Claude stable group | claude-opus-5 | Anthropic | $5 / $25 | see status page |
| Claude stable group | claude-sonnet-5 | Anthropic | $2 / $10 | see status page |
| Claude stable group | claude-opus-4-8 | Anthropic | $5 / $25 | see status page |
| Claude stable group | claude-sonnet-4-6 | Anthropic | $3 / $15 | see status page |
| Claude stable group | claude-haiku-4-5 | Anthropic | $1 / $5 | see status page |
| GPT group | gpt-5.6 | OpenAI | OpenAI list price | see status page |
| GPT group | gpt-5.5 | OpenAI | OpenAI list price | see status page |
| GPT group | gpt-5.2 | OpenAI | OpenAI list price | see status page |
All models with live status → · Updated: —
Pricing: top up and go, pay as you go
No subscription, no tiers: top up from ¥50, ¥1 buys $1 of credit, credits never expire.
- ✓Credits never expire, no daily cap
- ✓All public groups and models
- ✓Cache reads billed at cache-read price
- ✓$1 trial credit with LinuxDo sign-in
charge = list price × group multiplier × tokensShare of list price = multiplier × (¥1 / $1) ÷ 7.1 FX; plug in the group multiplier shown in the console.
List price $5 per million input tokens. One million input tokens cost $5 × the group multiplier in credit (¥1 = $1), versus about ¥35.5 buying direct. Output tokens likewise at $25 × the group multiplier.
Claude Code, Codex CLI or Cursor in 3 minutes
Create a key in the console after signing up, then change one BASE_URL.
export ANTHROPIC_BASE_URL=https://api.orcai.cc export ANTHROPIC_AUTH_TOKEN=sk-your-key claude --model claude-opus-5
# ~/.codex/config.toml model_provider = "orcai" model = "gpt-5.6" [model_providers.orcai] name = "OrcAI" base_url = "https://api.orcai.cc/openai/v1" wire_api = "responses" env_key = "ORCAI_API_KEY"
Base URL: https://api.orcai.cc/openai/v1 API Key: sk-your-key Model: claude-sonnet-5 / gpt-5.6 # Python from openai import OpenAI client = OpenAI(base_url="https://api.orcai.cc/openai/v1", api_key="sk-...")
Stable and secure by design
The pipeline and data handling are documented; everything here can be checked.
TLS 1.3 in transit
api.orcai.cc is HTTPS end to end; the website is fronted by Cloudflare.
Pool failover
A failing upstream account is rotated out automatically without touching your key.
No prompt storage
Usage logs hold billing fields only: token counts, model, duration. No prompt or completion bodies.
Probed every 90 s
Every model in every group is probed independently; results on the status page。
Frequently asked questions
Twelve questions on how relays work, pricing, model fidelity, setup and payment. More on the full FAQ page。
What is OrcAI?
OrcAI is a developer-focused AI model API relay: one Anthropic / OpenAI compatible endpoint (https://api.orcai.cc) for the Claude and GPT model families. As of September 2026 it serves 8,000+ registered developers and 300k+ requests per day.
What is an API relay and how is it different from the official API?
A relay is a gateway between you and the model vendor: you call a compatible endpoint with one key, the relay settles with its channels. The differences are payment (CNY / USDT, no overseas credit card), pricing (rate multipliers) and channel origin. OrcAI discloses every group's channel origin and never labels a non-official channel as official pass-through.
Which models are supported?
Claude Opus 5, Claude Sonnet 5, Claude Opus 4.8, Claude Sonnet 4.6, Claude Haiku 4.5; GPT-5.6 (sol / luna / terra), GPT-5.5, GPT-5.2. Model IDs match the vendors' IDs. The full list with live probe status is at https://orcai.cc/models/.
How is pricing calculated? What is a multiplier?
Charge = official unit price × group multiplier × tokens. Credits are in USD; ¥1 buys $1 of credit. Current group multipliers are shown in the console and may change with channel costs. Example: claude-opus-5 lists $5 per million input tokens, so one million input tokens cost $5 × the group multiplier in credit.
Why is the Claude stable group cheaper than list price? What is the channel?
The Claude stable group calls genuine Anthropic Claude models through a Kiro (AWS) licensed channel with prompt-cache optimisation (60%+ cache hits over the last 7 days). The GPT group uses a pool of official ChatGPT Pro accounts. Channel origins are shown in the console; OrcAI does not claim official Anthropic API pass-through.
Do credits expire? Is there a daily limit?
Credits never expire and there is no daily call cap. Groups may apply concurrency or rate limits to protect stability; you get a 429 and can retry.
How do I top up? Which payment methods are supported?
Open the Top Up page in the console: Alipay and USDT (TRC20 / ERC20 / BSC), minimum ¥50, credited 1:1 in USD. Signing in with a LinuxDo account adds $1 of trial credit.
How do I connect Claude Code, Codex CLI or Cursor?
Claude Code: set ANTHROPIC_BASE_URL=https://api.orcai.cc and ANTHROPIC_AUTH_TOKEN=your key. Codex CLI: point base_url in ~/.codex/config.toml to https://api.orcai.cc/openai/v1. Cursor or any OpenAI SDK: use https://api.orcai.cc/openai/v1 as the base URL. Step-by-step guides: https://orcai.cc/docs/.
Is it reachable from mainland China?
Yes. The gateway api.orcai.cc is reachable directly from mainland networks without a proxy; the website is served through Cloudflare.
How stable is it? Is there a status page?
Multi-account pools fail over automatically; every model in every group is probed every 90 seconds and latency / availability are published at https://orcai.cc/status/. 2.14 million requests were served in the last 7 days; the availability target is 99.9%.
How can I verify the relay is genuine and not swapping models?
Three checks: the model field in the response must equal the model you requested; usage token counts must match the text length; compare style and capability with the official model on the same prompt. In the last 7 days, none of the 1M+ OrcAI log entries carrying an upstream model identifier returned a model different from the one requested.
Do you log my prompts?
Gateway usage logs contain billing fields only: token counts, model, duration and timestamps. Request and response bodies are not stored. See https://orcai.cc/privacy/.
Start now and call Claude and GPT
with ¥1 = $1 credit
Sign up, create a key, top up from ¥50; credits never expire. LinuxDo sign-in adds $1 of trial credit.