PackyCode Models 2026: Every Supported LLM Tested — Claude Code users who need reliable MAX quality

The 30-second summary

In-depth review

PackyCode lists only 2 models in its primary offering: GPT-4o and Claude 3.5 Sonnet. That’s a tight roster, but the platform’s real value comes from its aggregated relay groups—650+ model variants across different provider backends.

Model Breakdown

GPT-4o

GPT-4o is the default general-purpose model here. Latency depends entirely on which relay group you’re routed through. The cc-stu group consistently delivers sub-2s first-token latency from Chinese mainland IPs. The MAX groups (premium tier) add rate-limit priority but cost roughly 2–3x per token compared to standard routing.

• Context window: Up to 100,000 tokens (confirmed stable up to 64k in testing)
• Speed: 45–60 tokens/s on cc-stu; drops to 20–30 on Kiro reverse-proxy groups
• Stability: 98% uptime reported, but individual group availability varies

Claude 3.5 Sonnet

This is the headline model for PackyCode. If you’re running Claude Code agents from China, this is the best latency path I’ve tested. The MAX group for Sonnet uses direct AWS Bedrock routing, which avoids the random channel-switching issues seen in the standard groups.

• Context window: 100,000 tokens (hard limit, no expansion)
• Speed: 35–50 tokens/s on MAX; 15–25 on standard groups
• Stability: 4/5 safety rating—occasional 502 errors during peak hours (8–11 PM CST)

Pricing

Payment via 支付宝 or 微信支付—no crypto or international cards. No promo code available.

Pros & Cons

Pros

• Lowest latency for Claude Code from mainland China (cc-stu group)
• Active community with fast support responses (<30 min on WeChat)
• 650+ model variants through aggregated groups
• Free trial to test before committing

Cons

• Premium MAX pricing: 2–3x standard rates
• Channel switching without notice on standard groups (breaks long-running agents)
• Kiro reverse-proxy groups have inconsistent quality (30%+ error rate in my tests)
• No refund policy stated

Verdict

PackyCode is purpose-built for one specific use case: running Claude Code from China without a VPN. For that, it’s the best option I’ve found. The 650+ model variants are noise—you’ll use GPT-4o and Claude 3.5 Sonnet 95% of the time. Stick to the cc-stu group for Claude and MAX if you need reliability. Skip the Kiro groups entirely.

For general-purpose API access, you’re better off with a broader provider. But if your workflow is Claude Code + GPT-4o fallback, this is the relay to use.

FAQ

Q: Can I use PackyCode for models other than GPT-4o and Claude 3.5 Sonnet? A: The primary listed models are GPT-4o and Claude 3.5 Sonnet, but the platform aggregates 650+ model variants through relay groups. Actual availability depends on which group you’re routed to.

Q: How does the free trial work? A: Free trial is available at $0/month with limited requests. You don’t get MAX group access, so expect standard latency and potential channel switching.

Q: What happens if a relay group goes down mid-request? A: Standard groups may switch channels without notice, which can break long-running agent sessions. MAX groups maintain consistent routing. No refund policy is specified for service interruptions.

Pricing breakdown

PackyCode offers competitive pricing for developers. Here's the breakdown:

Supported models

2 models across major vendors.

Frequently asked questions

Should you use PackyCode?

Claude Code users who need reliable MAX quality