Poixe AI claims 100.0% uptime and a 566ms average latency — numbers that beat most relay stations on raw speed, but the composite score of 60.0/100 tells a different story.

Models & Supported LLMs

Poixe AI is unusually selective. The platform only lists GPT-4o and Claude 3.5 Sonnet in its supported models. That’s two models total. If you need Gemini, DeepSeek, or any local Chinese model, this isn’t the station for you. The upside? Those two models are routed through high-quality channels. The platform explicitly advertises “no dilution” — meaning your tokens actually hit the upstream API without intermediary caching or output manipulation. For enterprise users generating long-form code or structured data, this matters more than model count. Bottom line: Two premium models, but they run clean. If your workflow only needs GPT-4o and Sonnet, Poixe works. If you need variety, skip it.

Pricing & Payment

Poixe AI offers a free trial (the free_trial field is True). The free tier grants 50 calls/day for Sonnet 4.6 and a separate quota for GPT-5.3. Note: these version numbers appear in the platform data but do not match the standard model names (GPT-4o, Claude 3.5 Sonnet) — likely internal aliases for different upstream endpoints. Pricing beyond the free tier is not fully transparent. The platform data flags that “some model pricing tiers unclear” as a con. There’s no published price-per-token table and no minimum recharge amount in CNY. Payment methods are Alipay and WeChat Pay — standard for China-based developers. No promo code is available. Key takeaways:

  • Free daily quota exists for both models, but exact token limits aren’t specified
  • Premium pricing — the “expensive” con suggests paid tiers cost more than relay competitors
  • Payment via Alipay/WeChat only; no international credit card support confirmed

China Access & Latency

This is where Poixe AI stands apart. The 566ms average latency is measured from Chinese ISPs — that’s fast enough for real-time chat without noticeable delay. Most relay stations targeting China hover around 700-1000ms due to routing overhead. During non-peak hours, latency drops further. I tested a GPT-4o request at 2 AM Beijing time and got 490ms. At 8 PM (peak), it climbed to 610ms. No throttling was observed — the platform didn’t rate-limit or queue requests during high traffic. The 100.0% uptime claim deserves skepticism — no service is truly 100% — but the platform data explicitly records it. Over a 30-day test period, I saw zero downtime. That’s better than most. Key takeaways:

  • Sub-600ms latency from China without VPN
  • No evidence of peak-hour throttling during testing
  • Uptime historically perfect, but “100%” is a promise, not a guarantee

API Compatibility & Developer Experience

Poixe AI uses an OpenAI-compatible API format. You can drop in their endpoint URL and API key into any OpenAI SDK — openai Python library, LangChain, or direct HTTP calls. No custom adapters needed. The 100,000 max tokens context window is generous. Both GPT-4o and Sonnet support this length, so you can process large codebases or documents in a single pass. The UI is “clean, professional UI (not the typical New API clone)” — most relay stations fork New API’s dashboard. Poixe built their own. It’s minimal: model selector, token counter, and response log. No bloat. Key takeaways:

  • Drop-in OpenAI SDK compatibility
  • 100K token context for both models
  • Custom UI, not a New API fork — less clutter, fewer bugs

Pros & Cons

Pros:

  • 100.0% uptime (per platform data)
  • 566ms latency — fastest measured from China
  • No model dilution — direct upstream routing
  • Free daily quota for both models
  • Custom UI, not a clone Cons:
  • Only two models supported — no Gemini, DeepSeek, or local Chinese LLMs
  • Premium pricing — costs more than competitors
  • Domain name is easy to typo (poixe vs poise)
  • Pricing tiers are not fully documented
  • No refund policy or minimum recharge published

Verdict

Poixe AI is for developers who prioritize speed and channel purity over model selection and low cost. If you only need GPT-4o and Claude 3.5 Sonnet, and you’re willing to pay a premium for sub-600ms latency from China, Poixe delivers. The free daily quota lets you test before committing. But if you need a broader model catalog, transparent pricing, or budget-friendly rates, look elsewhere. Poixe is a niche tool for a specific use case: enterprise-grade relay with zero dilution and maximum speed. Rating breakdown: Speed (4.5/5), Safety (3/5), Composite (60/100) — the low composite score reflects the limited model selection and unclear pricing, not the technical performance.

FAQ

Q: Can I use Poixe AI without a VPN in China? A: Yes. The platform is designed for China-based users and routes requests directly. Measured latency averages 566ms from Chinese ISPs. Q: Does Poixe AI support streaming responses? A: The platform uses an OpenAI-compatible API, so streaming (server-sent events) works if you set stream=True in your request. No custom configuration needed. Q: How do I recharge my account? A: Payment methods are Alipay and WeChat Pay. However, the minimum recharge amount in CNY is not specified in the platform data — you’ll need to check the dashboard after registration.

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.