Its free tier comes with access to GPT-5.5 and Claude Opus 4.6 — two models that aren’t even available on most newer relay stations.
Models & Pricing
The model list is short but specific. You get three options: GPT-5.5, Claude Opus 4.6, and Codex. No Gemini, no DeepSeek, no mid-tier models. This is a focused lineup aimed at developers who need top-end reasoning or code generation. Pricing follows a hybrid model. Standard usage is pay-per-call at ¥0.05 per request. The “MAX group” option targets Claude Opus 4.6 specifically — no dilution on Claude models, meaning your requests won’t get downgraded to cheaper variants. That matters when you’re debugging complex agents and need consistent behavior.
| Model | Pricing Model | Per-Call Cost |
|---|---|---|
| GPT-5.5 | Pay-per-call | ¥0.05 |
| Claude Opus 4.6 | Pay-per-call | ¥0.05 |
| Codex | Pay-per-call | ¥0.05 |
| MAX group (Claude) | Premium tier | Higher (not specified) |
| ¥1 minimum recharge keeps the barrier low. Payment methods are Alipay and WeChat Pay — standard for China, no international cards. |
Performance & Reliability
Uptime sits at 94.0%. That’s below the 99%+ you’d expect from established stations. This station launched in March 2026 — it’s new, and the uptime reflects growing pains. If you’re running production pipelines, this is a risk. For prototyping or personal projects, it’s workable. Safety rating is 4/5. No filter details are provided, but that score suggests minimal content blocking for Chinese developers accessing Western models.
API Compatibility
No explicit API compatibility info is given. Given the model names (GPT-5.5, Codex), expect OpenAI-style endpoint formatting. Claude Opus 4.6 likely uses Anthropic-style requests. Without documentation confirmation, you’ll need to test endpoint paths manually.
Pros & Cons
Pros
- Free trial gives access to GPT-5.5 and Claude Opus 4.6 — no credit card needed
- MAX group avoids Claude model dilution: your request hits the exact model you paid for
- Codex is bundled with GPT-5.5 access — useful for code generation workflows
- ¥1 minimum recharge — lower than most stations requiring ¥10+ Cons
- Launched March 2026 — less than one month of operational history
- 94.0% uptime is unreliable for production workloads
- No refund policy stated — risk if you prepay and the station goes down
- Limited model selection: no Gemini, no DeepSeek, no mid-range options
- MAX group pricing is relatively high — no exact numbers listed
Verdict
Faxian AI is a gamble worth taking only if you’re testing the newest models. The free trial gets you GPT-5.5 and Claude Opus 4.6 without spending a cent — that’s rare. But 94% uptime and zero track record mean you shouldn’t depend on it for anything critical. Use it for evaluation: spin up a prototype, test model behavior, then decide if you want to commit to a more reliable station. Don’t wire it into production yet.
FAQ
Can I use Faxian AI without a VPN in China? Yes. This is a relay station designed for Chinese developers — no VPN required to access GPT-5.5, Claude Opus 4.6, or Codex. What happens if I use up my free trial? You can recharge with ¥1 minimum via Alipay or WeChat Pay. Pricing is ¥0.05 per call after the trial. Does Faxian AI support streaming responses? Not indicated in the available data. You’ll need to test streaming endpoints manually or contact support for API documentation.
Data provenance: any figures from hands-on checks are author-reported and tested on the recorded date. Independent verification is unavailable unless a source is linked.
