CCFly lists exactly 2 models: GPT-4o and Claude 3.5 Sonnet. That’s it. No Gemini, no DeepSeek, no Mistral. This is a focused, Claude-first relay station aimed at developers who want premium Claude access without VPN.

Model Breakdown

Claude 3.5 Sonnet

This is CCFly’s flagship offering. The platform data shows a 100.0% uptime record and 514ms latency, which puts it in the fast lane for Claude access. The safety rating sits at 4/5 — not perfect, but solid. The context window maxes out at 100,000 tokens, which is the standard for Claude 3.5 Sonnet. You won’t hit the full 200K limit here, but 100K is enough for most code review, long document analysis, and multi-file refactoring tasks. Key takeaways:

  • 100% uptime means you won’t get dropped mid-session
  • 514ms latency is competitive for Claude relay stations
  • 100K context window covers most production use cases

GPT-4o

CCFly includes GPT-4o but it’s clearly secondary. The platform lists it, but the entire service design — MAX carpool, quality focus — centers on Claude. GPT-4o is here for fallback or variety, not as a primary workhorse. You get the same 100K context window and same uptime guarantees. But don’t expect multi-model optimization. This is a Claude box that also runs GPT-4o. Key takeaways:

  • GPT-4o is present but not optimized
  • Same uptime and latency as Claude
  • Use it for tasks where GPT-4o outperforms Claude (structured data extraction, JSON mode)

Stability and Incident Patterns

Generic relay reviews skip per-model SLA. Here’s what matters: CCFly’s 100.0% uptime across both models suggests they don’t prioritize one over the other during outages. When incidents hit, both models likely degrade together. The speed rating of 4.5/5 backs this up. You’re not getting throttled differently per model. That’s rare — most relay stations have hidden rate limits on cheaper models. Bottom line: If Claude goes down, GPT-4o goes down too. No preferential treatment.

Context Window Limits

Both models share the 100,000 token cap. This is below Claude’s native 200K limit but matches what most relay stations offer at this price point. For code generation, debugging, and documentation, 100K is sufficient. For full-codebase analysis, you’ll hit the wall. No tiered context windows exist here. One cap, both models.

Price vs. Official Access

CCFly runs a MAX carpool service for Claude Code — shared exclusive accounts at ¥2300/month solo. That’s more expensive than a direct Claude Pro subscription (which costs about $20/month or ~¥140). You’re paying for the relay infrastructure and China access, not for model access itself. The platform lists ¥0/month as base price with no free trial. That likely means pay-as-you-go or recharge-based billing. Payment methods are Alipay and WeChat Pay — standard for Chinese developers.

Pros & Cons

Pros:

  • Claude-focused expertise with dedicated infrastructure
  • 100% uptime across both models
  • 514ms latency is fast for a relay station
  • Alipay/WeChat Pay accepted
  • Speed rating 4.5/5 — consistent performance Cons:
  • Only 2 models — no Gemini, DeepSeek, Mistral, or others
  • MAX carpool ¥2300/month is expensive vs. official pricing
  • 100K context window below Claude’s native 200K
  • No promo code or refund policy specified
  • Multi-model users will need additional providers

Verdict

CCFly is for one type of developer: someone who needs reliable, fast Claude 3.5 Sonnet access in China without VPN, and is willing to pay a premium for it. If you’re building Claude-heavy workflows (code generation, long-form analysis, agentic tasks), this delivers. If you need model diversity or lower prices, look elsewhere. The 80.0/100 composite score reflects the narrow focus. High marks for reliability and speed, lower for model breadth and pricing.

FAQ

Does CCFly support DeepSeek or Gemini?

No. The platform only supports GPT-4o and Claude 3.5 Sonnet. No other models are available.

What is the MAX carpool service?

It’s a shared exclusive account system for Claude Code, priced at ¥2300/month solo. You get a dedicated account shared among a small pool of users, with priority access.

Can I use CCFly without recharging?

The base price is ¥0/month, but there’s no free trial. You’ll need to recharge via Alipay or WeChat Pay to access the models. Minimum recharge amount is not specified.

What happens if Claude 3.5 Sonnet goes down?

Both models share the same infrastructure. If Claude goes down, GPT-4o likely goes down too. The 100% uptime record suggests this is rare.

Is CCFly suitable for multi-model workflows?

Not really. With only 2 models, you’ll need additional relay stations for Gemini, DeepSeek, or other models. CCFly is best as a dedicated Claude access point.

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.