GPTZZZ: one API account for GPT, Claude, Gemini and more
Updated
GPTZZZ is an independent third-party AI API relay. One account and one prepaid balance give you models from seven vendors. OpenAI groups use an OpenAI-compatible base URL, Claude groups use the Claude Messages endpoint, and for the other groups the console and the API docs show which endpoints to use. An individual operator runs it. It is not affiliated with OpenAI, Anthropic, Google, DeepSeek, xAI, Alibaba or Xiaomi. The API opened on 9 September 2026. The service is built for Chinese-speaking developers: top-ups are in CNY via Alipay, and the documentation and support pages are in Chinese.
At a glance
| Item | Value |
|---|---|
| OpenAI-compatible base URL | https://gptzzz.ai/v1 |
| Claude Messages endpoint | https://gptzzz.ai/v1/messages |
| Claude Code base URL | https://gptzzz.ai (without /v1) |
| Authentication | Authorization: Bearer, x-api-key or x-goog-api-key. Keys in the query string are rejected. |
| Catalogue | 48 models, 7 vendors, 10 billing groups (pricing page, updated 30 September 2026) |
| Prices | CNY per 1M tokens, group multiplier already applied |
| Top-up | ¥1 = $1 of balance, Alipay, ¥10 to ¥2,000 per top-up, no fee |
| Operator | Independent, individually operated; not affiliated with any model vendor |
Endpoints and authentication
OpenAI-compatible clients use the base URL https://gptzzz.ai/v1. Your client adds /chat/completions or /responses itself. Do not append /v1 twice.
For the OpenAI groups, the operator lists these endpoints as tested: /v1/models, /v1/chat/completions (including streaming, tool calls and JSON mode) and /v1/responses. Send image input as base64. Embeddings are not supported.
Claude groups use the Claude Messages API at https://gptzzz.ai/v1/messages. For the other groups, check the console and the API docs (Chinese) for the available endpoints.
Send your key in an Authorization: Bearer, x-api-key or x-goog-api-key header. Keys passed in the query string are rejected. Keep keys on the server or in local environment variables, never in front-end code.
Models
As listed on the pricing page (Chinese), updated 30 September 2026: 48 models from 7 vendors in 10 billing groups.
| Vendor | Models | Billing groups |
|---|---|---|
| OpenAI GPT | 20 | opai puls, opai pro, pro200-500不降智, GPT官key |
| Anthropic Claude | 12 | Claude官key |
| Google Gemini | 6 | Gemini官key |
| Xiaomi MiMo | 4 | MiMo 小米官key |
| DeepSeek | 2 | DeepSeek官key |
| xAI Grok | 2 | Grok官key |
| Qwen (Alibaba) | 2 | Qwen 通义官key |
There are also five gpt-image models. They cost ¥0.20 per image and are only in the opai puls and opai pro groups.
Use the exact model ID from the pricing page or the console, for example gpt-5.5, claude-sonnet-5, gemini-3.8-flash or deepseek-v4.1-flash. GPTZZZ IDs can differ from the vendor's own IDs.
Billing groups and API keys
Each API key belongs to exactly one billing group. A key can only call the models in its group. You pick the group when you create the key in the console (/keys). To use several vendors, create one key per group. All keys draw on the same balance.
The pricing page lists two types of group:
- GPT subscription-pool groups (GPT 订阅号池): opai puls (0.05x), opai pro (0.1x) and pro200-500不降智 (0.25x). They are billed at the official price times the multiplier.
- Official-key groups (官方 Key): Claude官key (0.39x), GPT官key (0.54x), MiMo 小米官key (0.64x), Grok官key (0.67x), Qwen 通义官key (0.7x), DeepSeek官key (0.77x) and Gemini官key (0.92x). Each is billed at its own price card times the multiplier.
The three cheapest GPT groups are subscription-pool groups, not official-key groups. For GPT models, only GPT官key uses official API keys. Choose the group that fits your workload before you build on it.
A multiplier only means something inside its own group. To compare groups, compare the listed prices. The models also differ by group; the group table on the pricing page shows which models each group has.
Prices
Prices are in CNY per 1M tokens and already include the group multiplier. A ¥1 top-up adds $1 of balance, so the listed number is what you pay in CNY. Examples from the pricing page:
| Model | Group | Input | Output | Cache read |
|---|---|---|---|---|
gpt-5.5 | opai puls (0.05x) | ¥0.25 | ¥1.50 | ¥0.025 |
gpt-5.5 | opai pro (0.1x) | ¥0.50 | ¥3.00 | ¥0.05 |
gpt-5.5 | pro200-500不降智 (0.25x) | ¥1.25 | ¥7.50 | ¥0.13 |
gpt-5.5 | GPT官key (0.54x) | ¥3.91 | ¥23.40 | ¥0.39 |
claude-opus-5.5 | Claude官key (0.39x) | ¥18.70 | ¥93.70 | ¥0.94 |
The pricing page lists the official prices as US$5.00 / $30.00 for gpt-5.5 and US$4.00 / $20.00 for claude-opus-5.5 (input / output per 1M tokens). It also shows a 元/刀 figure (CNY per US$1 of official-price usage): the site's CNY input price divided by the official USD input price. For gpt-5.5 in opai puls that is ¥0.25 ÷ US$5.00 = ¥0.05. For claude-opus-5.5 it is about ¥4.68.
Some GPT models bill the whole request at a higher rate when the input is over 272K tokens. For gpt-5.5 in opai puls that rate is ¥0.50 input and ¥2.25 output. Cache writes and long-context rates are listed per model on the pricing page. If the console shows a different price, the console price applies.
Top-up and payment
- Usage is deducted from a prepaid balance. ¥1 adds $1 of balance.
- Payment method: Alipay. Each top-up is ¥10 to ¥2,000, with no fee.
- Unused balance can be refunded on request at any time, through [email protected] or Telegram.
- Invoices are not issued, because the service is individually operated.
- There is no sign-up bonus.
Quick start
- Create an account at /register.
- Top up your balance with Alipay.
- In the console (/keys), create an API key for the group that contains your model.
- Point your client at the base URL and send one small non-streaming request. Add streaming, tools or long context after that works.
curl (Chat Completions)
gpt-5.5 is in all four GPT groups, so a key from any GPT group works here.
export GPTZZZ_API_KEY="<your GPT group key>"
curl https://gptzzz.ai/v1/chat/completions \
-H "Authorization: Bearer $GPTZZZ_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-5.5", "messages": [{"role": "user", "content": "Hello"}]}'Python (openai SDK)
import os
from openai import OpenAI
client = OpenAI(
base_url="https://gptzzz.ai/v1",
api_key=os.environ["GPTZZZ_API_KEY"],
)
resp = client.chat.completions.create(
model="gpt-5.5",
messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)
# Responses API
resp = client.responses.create(model="gpt-5.5", input="Hello")
print(resp.output_text)Claude Messages API
Use a key from the Claude官key group. A GPT-group key cannot call Claude models, so set the variable again.
export GPTZZZ_API_KEY="<your Claude官key group key>"
curl https://gptzzz.ai/v1/messages \
-H "x-api-key: $GPTZZZ_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{"model": "claude-sonnet-5", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'With the Anthropic Python SDK, set base_url to the host only. The SDK adds /v1/messages itself.
import os
from anthropic import Anthropic
client = Anthropic(
base_url="https://gptzzz.ai",
api_key=os.environ["GPTZZZ_API_KEY"],
)
msg = client.messages.create(
model="claude-sonnet-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello"}],
)
for block in msg.content:
if block.type == "text":
print(block.text)Claude Sonnet 5 has thinking on by default. The response can start with a thinking block, so print only the text blocks. Thinking tokens count toward max_tokens, so do not set it too low. See Anthropic's thinking documentation.
Claude Code
Use a key from the Claude官key group. ANTHROPIC_BASE_URL has no /v1.
export GPTZZZ_API_KEY="<your Claude官key group key>"
export ANTHROPIC_BASE_URL="https://gptzzz.ai"
export ANTHROPIC_AUTH_TOKEN="$GPTZZZ_API_KEY"
export ANTHROPIC_MODEL="claude-sonnet-5"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-sonnet-5"
claudeClaude Code sends ANTHROPIC_AUTH_TOKEN as Authorization: Bearer. ANTHROPIC_API_KEY, sent as x-api-key, also works. ANTHROPIC_DEFAULT_HAIKU_MODEL sets the model behind the haiku alias, which Claude Code also uses for background tasks, so set it to a model ID in your group. See the Claude Code environment variable reference.
Desktop client
The GPTZZZ desktop client is at version 3.21.0, released on 12 September 2026. It runs on macOS (one universal .dmg for Apple silicon and Intel) and on Windows x64 and ARM64 (.msi installer or portable .zip).
It works with five tools: Claude Code, Codex, Grok and two tools that the Chinese release notes call 小龙虾 and 爱马仕. After you sign in, it adds GPTZZZ providers to the provider list. It does not switch to them automatically. You can restore your original configuration in one click. Downloads: desktop client page (Chinese).
Language and support
- The documentation, pricing, status and support pages are in Chinese.
- Email: [email protected]. Telegram: @gptzzz1. QQ group: 103966538.
- For account and balance questions, sign in to the console first.
- Never send a full API key. A masked prefix and suffix, such as
sk-abcd…wxyz, is enough. If a key has leaked, revoke it in the console and create a new one. - The status page (Chinese) shows a public health-check signal. It is not an SLA and does not cover every model.
Code and integrations
- github.com/gptzzz: example code, the Dify plugin source and an OpenAI-compatible API checker.
- GPTZZZ plugin on the Dify Marketplace (model provider).
- LangChain Python docs: GPTZZZ is listed in the chat model table.
- Setup guides (Chinese): Claude Code, Codex CLI, Vercel AI SDK, Dify, API error codes.
Common questions
Is GPTZZZ an official OpenAI, Anthropic or Google service?
No. GPTZZZ is an independent third-party API relay run by an individual operator. It is not affiliated with OpenAI, Anthropic, Google or any other model vendor.
Can one API key call every model?
No. Each key belongs to one billing group and can only call the models in that group. Create one key per group in the console. All keys draw on the same balance.
Which GPT group uses official OpenAI API keys?
GPT官key. The pricing page lists it as an official-key group. The cheaper GPT groups, opai puls, opai pro and pro200-500不降智, are listed as GPT subscription-pool groups.
How do I pay, and can I get a refund?
Top-ups are paid in CNY with Alipay, from ¥10 to ¥2,000 each, with no fee. ¥1 adds $1 of balance. Unused balance can be refunded on request through [email protected] or Telegram. Invoices are not issued, and there is no sign-up bonus.
Is the documentation available in English?
No. The documentation, pricing, status and support pages are in Chinese. This page summarises the main facts in English.