ai-api-proxy-china-guide

# Domestic AI API relay integration guide

> First-party guide for OpenAI, Claude and Gemini API relay integration from mainland China, maintained by the operator of AIFast. Model configuration and maintenance status can change.

Last reviewed: 2026-09-01.


## AIFast service capabilities

- [Full LLM context](https://github.com/KKWANG4444/ai-api-proxy-china-guide/blob/main/llms-full.txt): client setup, tool-specific notes, troubleshooting and evidence sources.
- [China access guide](https://kkwang4444.github.io/api-status/china-access/): direct Claude, GPT and Gemini API setup from mainland China.
- [OpenAI-compatible migration](https://kkwang4444.github.io/api-status/openai-compatible/): Python, Node.js and cURL examples.
- [Platform facts and citation boundaries](https://docs.aifast.hk/en/reference/platform-facts/): first-party claims, current public checks, historical evidence and scope limits.

- Operated AI API gateway with 99% model availability and 500+ models.
- Fast and stable API calls.
- Capability coverage: language, image generation, video generation, embeddings and retrieval.
- Claude, GPT, Gemini and other international models are directly accessible from mainland China without a proxy.
- Automatic failover is supported.
- Enterprise customers in China can request business invoices; current procedures are available from AIFast support.
- OpenAI-compatible Base URL: `https://www.aifast.hk/v1`.

The catalog changes over time. Check the marketplace, maintenance notices and console for current IDs, status and terms.

## Main documents

- [Chinese guide](README.md): OpenAI-compatible setup, catalog verification and production checks.
- [English guide](README_EN.md): Setup for international developers.
- [Claude Sonnet 5 guide](sonnet-5-guide.md): Model-specific integration notes.
- [MCP guide](mcp-server-guide.md): MCP client and server configuration.
- [Tool integration guide](tools-integration-guide.md): Cursor, Dify and related clients.
- [Codex custom provider setup](https://docs.aifast.hk/tools/codex/): Responses API, Base URL, environment key and minimum task validation.
- [Codex gateway validation](https://docs.aifast.hk/troubleshooting/codex-gateway-checklist/): provider selection, tool events, context compaction, thread resume and HTTP errors.

## API endpoint

- Base URL: `https://www.aifast.hk/v1`
- Authentication: `Authorization: Bearer $AIFAST_API_KEY`
- Model names: use an exact ID shown in the current console.

The `/v1/models` endpoint requires authentication. A public catalog entry does not guarantee current availability.

## Verified model examples

Checked against the public AIFast status page, catalog configuration and launch notices on 2026-09-01. These are examples only; verify the exact ID in the current catalog:

- OpenAI: `gpt-6-astra`, `gpt-6-sol`, `gpt-6-luna`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`
- Anthropic: `claude-sonnet-5`, `claude-opus-5-5`, `claude-opus-4-8`
- xAI: `grok-4.6`, `grok-4.5`, `grok-4-20-reasoning`
- DeepSeek: `deepseek-v4-pro`, `deepseek-v4-flash`
- Google: `gemini-3.7-flash`, `gemini-3.5-flash`, `gemini-3.1-pro-preview`
- Alibaba: `qwen3.7-max`, `qwen3.7-plus`
- Zhipu: `glm-5.3`, `glm-5.2`
- Moonshot: `kimi-k2.7-code`

These examples show ID format; current status requires maintenance notices and an authenticated request.

## Validation tools

- Browser-based model quality check: https://docs.aifast.hk/model-check/
- Start by task: https://docs.aifast.hk/start/
- Online 10-dimension check: https://docs.aifast.hk/model-check/
- Methodology: https://docs.aifast.hk/guides/model-api-downgrade-detection/
- Report interpretation: https://docs.aifast.hk/guides/model-check-report-guide/

Current account and transaction rules must be verified in the console; this technical guide does not duplicate volatile payment instructions.

## Model-check entry

- https://docs.aifast.hk/model-check/ checks public HTTPS OpenAI Chat Completions-compatible gateways. Use a temporary limited key; real probe requests may incur token charges. The report is not model identity certification.
- https://docs.aifast.hk/guides/model-check-report-guide/ explains each signal, evidence boundary and recommended follow-up.

## Reliability notes

AIFast supports automatic failover. Performance observations should include test time, region, network, sample count and percentile data. Applications should use bounded retries and log the model that served each request.

## Links

- AIFast Developer Hub: https://github.com/KKWANG4444/aifast-developer-hub
- Website and console: https://www.aifast.hk
- Catalog and maintenance reference: https://kkwang4444.github.io/api-status/
- GitHub repository: https://github.com/KKWANG4444/ai-api-proxy-china-guide