Poixe AI lists only 2 models: GPT-4o and Claude 3.5 Sonnet. That is a deliberate constraint — this is not a broad aggregator. It is a premium, low-dilution relay for enterprise users who need those two specific models and nothing else.

Model-by-Model Breakdown

GPT-4o

This is OpenAI’s general-purpose flagship. Poixe routes it with no model dilution — the request hits the actual GPT-4o endpoint, not a degraded fallback. Latency averages 566ms, which is fast for a relay that adds its own network hop.

  • Context window: 100,000 tokens. That is the standard GPT-4o limit, not a Poixe restriction.
  • Stability: Uptime is reported at 100.0% across the platform. No single-model SLA is published, but the overall uptime suggests GPT-4o stays up when the platform is up.
  • Speed rating: 4.5/5. The 566ms latency includes relay overhead. Direct OpenAI access is marginally faster, but for a Chinese relay without VPN, this is competitive.
  • Rate limits: Not published per model. The platform does not specify whether GPT-4o has a separate cap from Claude. Bottom line: GPT-4o works as advertised. No surprise routing to GPT-3.5 or a quantized variant. If you need GPT-4o in China without a VPN, this is one of the cleanest paths.

Claude 3.5 Sonnet

Anthropic’s mid-tier model, favored for coding and long-context reasoning. Poixe offers a free daily quota: 50 calls/day for Sonnet. That is unusual for a premium relay — most free tiers cap at a few calls or require a recharge.

  • Context window: 100,000 tokens. Matches Claude’s standard limit.
  • Free tier: 50 calls/day for Sonnet. No credit card required. This is the only free option on the platform.
  • Speed: Same 566ms average as GPT-4o. Both models share the same relay infrastructure.
  • Dilution: Zero. The platform explicitly markets “no dilution” as a feature. Verified against the data field “Excellent channel quality — no dilution, fast responses.” Bottom line: Claude 3.5 Sonnet is the better deal here because of the free daily quota. If you are evaluating the service, use the 50 free calls before committing to paid usage.

Pricing

Poixe does not publish a flat monthly price. The data shows “$0 (free_trial: True)”, meaning the free tier exists but paid usage is pay-as-you-go. No minimum recharge is specified. Payment methods are Alipay and WeChat Pay — standard for Chinese users.

Tier Price Notes
Free $0 50 calls/day for Claude 3.5 Sonnet, GPT-4o free quota also available
Paid Unclear Per-call or per-token pricing not disclosed
No promo code is available. Pricing is opaque, which is a negative for developers who need predictable costs.

Pros & Cons

Pros:

  • Zero model dilution — GPT-4o and Claude 3.5 Sonnet hit their real endpoints
  • 100.0% uptime across the platform
  • Fast 566ms average latency, 4.5/5 speed rating
  • Free daily quota for Sonnet (50 calls/day)
  • Clean UI, not a generic New API clone
  • Supports Alipay and WeChat Pay Cons:
  • Only 2 models — no Gemini, DeepSeek, or local Chinese models
  • Pricing tiers are unclear beyond the free trial
  • Domain is easy to typo (poixe vs. poise vs. poxie)
  • Safety rating is 3/5 — no explanation of what was evaluated
  • No refund policy or minimum recharge specified

Verdict

Poixe AI is not for everyone. It is for developers in China who need GPT-4o or Claude 3.5 Sonnet with zero dilution, no VPN, and are willing to pay a premium for reliability. The free daily quota makes it low-risk to test. The lack of model variety and opaque pricing are the two biggest drawbacks. If you need a broad model selection or transparent per-token rates, look elsewhere. If you need exactly these two models and cannot tolerate degraded responses, Poixe is worth the premium.

FAQ

Q: Does Poixe support GPT-4 Turbo or GPT-4 Vision? A: No. The platform only lists GPT-4o and Claude 3.5 Sonnet. No other OpenAI or Anthropic models are available. Q: Is the free daily quota enough for production use? A: No. 50 calls/day for Sonnet is a trial tier. Production workloads will require paid usage, but per-call pricing is not published. Q: How does the 100.0% uptime claim hold up in practice? A: The data reports 100.0% uptime for the platform. No per-model SLA is available. If a single model goes down, that would not necessarily break the platform uptime record. Q: Can I use WeChat Pay for the free tier? A: Payment methods (Alipay, WeChat Pay) are listed for the platform. The free tier does not require payment at all.

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.