PackyCode lists 2 models: GPT-4o and Claude 3.5 Sonnet. If you need Claude Code access from China with low latency, this is the site to look at.

Model Availability & Coverage

PackyCode’s model selection is deliberately narrow — they focus on quality over quantity. The two models supported are GPT-4o and Claude 3.5 Sonnet. The platform claims to support 650+ model variants through its group system, but the primary advertised endpoints are these two. Bottom line: If you need GPT-4o or Claude 3.5 Sonnet, PackyCode covers them. If you need Gemini, DeepSeek, or any other model, look elsewhere.

Performance Benchmarks

Latency

The platform reports 172ms average latency. For Claude Code users in China, this is the best available option according to the data. Most relay stations serving Chinese developers sit in the 200-300ms range, so 172ms is notably faster.

Speed Rating

Speed rating is 5.0/5 — the maximum. This aligns with the latency figure. The platform prioritizes MAX quality groups, which typically use direct API connections rather than cascading relays.

Uptime

98.0% uptime. This is acceptable but not exceptional. For comparison, most top-tier relay stations target 99.5%+. The 2% downtime window means you’ll encounter roughly 14.4 minutes of downtime per day on average. Key takeaways:

  • 172ms latency is best-in-class for Claude Code from China
  • 98.0% uptime is below the 99.5%+ standard
  • Speed rating is perfect, but that measures throughput not reliability

Context Window & Token Limits

Max tokens: 100,000. This applies across both models. For Claude 3.5 Sonnet, this is below the native 200K context window. For GPT-4o, this matches the standard 128K limit fairly closely. This is a per-request cap, not a rate limit. The data doesn’t specify whether there are per-minute or per-day rate limits on different tiers. Key takeaways:

  • 100K context window is sufficient for most code analysis tasks
  • Claude 3.5 Sonnet is capped below its native 200K capability
  • No rate limit data available — assume standard API throttling

Pricing & Payment

Plan Price Notes
Free trial $0 Available, limited usage
Standard Not specified No pricing data provided
MAX groups Premium Higher quality, higher cost
Payment methods: Alipay and WeChat Pay. This is standard for Chinese users. No promo code is available.
The platform has a free trial option, which lets you test latency and model quality before committing. The “cc-stu group” is mentioned as offering good quality at fair price — this appears to be a community-managed group within the platform.
Key takeaways:
  • Free trial available for testing
  • Chinese payment methods only (Alipay, WeChat Pay)
  • No refund policy specified — assume no refunds

Pros & Cons

Pros:

  • Best latency for Claude Code from China (172ms)
  • Active community with fast support
  • cc-stu group offers good quality at fair price
  • Supports 650+ model variants through group system
  • Free trial available Cons:
  • Premium pricing on MAX groups
  • Some groups may switch channels without notice
  • Kiro reverse-proxy groups have inconsistent quality
  • Only 2 models officially listed (GPT-4o, Claude 3.5 Sonnet)
  • No refund policy specified
  • Uptime (98.0%) is below industry standard

Verdict

PackyCode is a specialized tool for a specific use case: Claude Code users in China who need the lowest possible latency. The 172ms latency is the standout feature. If you’re primarily using Claude 3.5 Sonnet or GPT-4o for code generation and analysis, this platform delivers. The tradeoffs are clear: limited model selection, 98.0% uptime, and premium pricing for the best groups. The inconsistent quality of Kiro reverse-proxy groups and potential channel switching without notice are real risks. For developers who need model variety or higher uptime guarantees, look at platforms with broader model support and published SLAs. For those who need fast Claude Code access from China and are willing to accept the limitations, PackyCode is worth the free trial.

FAQ

Q: Does PackyCode support DeepSeek or Gemini models? A: No. The platform only officially lists GPT-4o and Claude 3.5 Sonnet. The “650+ model variants” claim refers to group configurations, not distinct model endpoints. Q: What is the actual context window for Claude 3.5 Sonnet on PackyCode? A: The max tokens is 100,000 across all models. For Claude 3.5 Sonnet, this is a reduction from the native 200K context window. Q: Can I get a refund if the service is unreliable? A: The platform data does not specify a refund policy. Assume purchases are final. The free trial is available to test service quality before committing funds.

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.