Evaluate first
- LiteLLM 84
- OpenRouter 82
- 七牛云 AI 推理 API 80
- Portkey 78
- New API 76
Trusted Directory
Claude API Relay Providers and Risk Comparison
Claude API relay demand often comes from Claude Code, custom tooling, and regional access constraints. This page focuses on providers that claim Claude support and highlights code-context privacy risk.
Quick Take
- For Claude Code and coding workflows, privacy and prompt logging matter more than headline price.
- Cheap Claude relay access needs extra verification for model consistency and context length.
- Do not route company code through an unverified relay.
Provider Groups
Test with small amount
- 302.ai 68
- APIPod 68
- CloseAI 66
Use caution
No matching providers yet
Provider Comparison
| Provider | Models | Tools | Payment | Min | Invoice | Risk | Verified |
|---|---|---|---|---|---|---|---|
| LiteLLM | OpenAI / Claude / Gemini / Grok | OpenAI SDK / LangChain / LlamaIndex | Self-hosted | No minimum for self-hosting | Unknown | Low risk · 84 | 2026-05-18 |
| OpenRouter | OpenAI / Claude / Gemini / Grok | OpenAI SDK / LiteLLM / LangChain | Credit card | $5 | Unknown | Low risk · 82 | 2026-05-18 |
| 七牛云 AI 推理 API | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / Claude Code | Alipay / WeChat Pay / Enterprise transfer | Pay as you go | Supported | Low risk · 80 | 2026-05-18 |
| Portkey | OpenAI / Claude / Gemini | OpenAI SDK / LiteLLM / LangChain | Credit card | No minimum | Supported | Low risk · 78 | 2026-05-18 |
| New API | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / LiteLLM / LangChain | Self-hosted | No minimum for self-hosting | Unknown | Low risk · 76 | 2026-05-18 |
| OfoxAI | OpenAI / Claude / Gemini / DeepSeek | Claude Code / OpenClaw / Cursor | Alipay / WeChat Pay / USDT | ¥50 | Unknown | Medium risk · 72 | 2026-05-18 |
| 柏拉图AI | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / Cursor | Alipay / WeChat Pay / Enterprise transfer | Enterprise minimum | Supported | Medium risk · 72 | 2026-05-18 |
| AiHubMix | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / Cursor / LiteLLM | Alipay / WeChat Pay | ¥100 | Supported | Medium risk · 70 | 2026-05-18 |
| 接口AI | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / Claude Code / Cursor | Alipay / WeChat Pay | ¥100 | Supported | Medium risk · 70 | 2026-05-18 |
| 302.ai | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / Cursor / LiteLLM | Alipay / WeChat Pay | ¥50 | Unknown | Medium risk · 68 | 2026-05-18 |
| APIPod | OpenAI / Claude / Gemini / DeepSeek | OpenAI SDK / LiteLLM / LangChain | Unknown | Unknown | Unknown | Medium risk · 68 | 2026-05-18 |
| CloseAI | OpenAI / Claude | OpenAI SDK / Cursor | Alipay / WeChat Pay | ¥100 | Supported | Medium risk · 66 | 2026-05-18 |
How to Choose
Check model support, API compatibility, Claude Code guides, privacy policy, logging behavior, and verification date before using a Claude relay provider.
Risk Notes
Claude relay risks include code-context exposure, model substitution, shortened context windows, unclear rate limits, and balance disputes.
The trust score is not a purchase recommendation and does not guarantee service availability. Test with small amounts first. For sensitive code, customer data, or production workloads, prefer official APIs, BYOK, or enterprise-compliant options.
FAQ
Can a Claude API relay work with Claude Code?
Some providers claim Claude Code support, but you should test Base URL, model names, tool use, and streaming output with a small amount first.
Can a Claude relay expose my code?
Any third-party relay can potentially observe prompts and code context. Sensitive repositories should use official APIs or compliant enterprise options.
How can I detect Claude model substitution?
Compare fixed prompts, long-context behavior, tool calls, and structured JSON output against official API behavior.
Are cheap Claude API relays reliable?
Low price is not proof of abuse, but extremely low price is a risk signal that needs extra verification.
Submit or Correct Data
If provider data is outdated, missing risk notes, or a new provider should be listed, send feedback.
Send feedback