LinkAPI’s composite score of 57.6/100 and 96% uptime put it in the mid-tier for AI relay stations — usable, but not the benchmark you’d bet your production pipeline on.
Models & Access
The platform explicitly lists GPT-4o and Claude 3.5 Sonnet as supported models. But here’s where it gets interesting: the “Pros” field mentions support for Gemini, DeepSeek, Kimi, Grok, GLM, and Qwen too. That’s a wider net than most single-provider relays cast. For Chinese developers, this breadth matters. You get Claude for creative tasks, GPT-4o for structured output, and domestic models like GLM/Qwen when you need lower latency from China-based endpoints. No VPN required — that’s the whole point. Reality check: the “widest model support” claim needs a qualifier. The cheapest channel “consistently dilutes” — meaning you might request GPT-4o but get routed to a lower-quality variant if you pick the budget option. If you need deterministic model quality, pay-per-token and explicitly select the premium channel.
Pricing & Payment
| Feature | Detail |
|---|---|
| Monthly cost | $0 (no free trial) |
| Minimum recharge | ¥1 (about $0.14) |
| Payment methods | Alipay, WeChat Pay |
| Pricing models | Pay-per-call + pay-per-token |
| Promo code | None |
| ¥1 minimum recharge is the lowest barrier I’ve seen. Most relays require ¥10-20 minimum. This makes LinkAPI viable for testing without commitment — throw ¥5 at it, run 50-100 calls, decide if the quality holds up. | |
| The dual pricing model (per-call vs per-token) gives flexibility. Use pay-per-call for short, predictable prompts; pay-per-token for long-context work where you’d get gouged on per-call pricing. | |
| No promo code available. That’s fine — at ¥1 minimum, you don’t need one. |
Uptime & Reliability
96% uptime over whatever measurement period they use. That’s roughly 14 hours of downtime per month. For a hobby project or personal assistant bot, that’s survivable. For a customer-facing SaaS with SLA requirements? You’ll want a fallback relay. The 3/5 safety rating suggests occasional routing issues or model mismatches. Combined with the “variable quality day-to-day” con, expect inconsistency on the cheapest tier. Test during Chinese peak hours (8-11 PM CST) to see worst-case behavior. 100K max tokens is generous. That covers most long-context tasks — codebase analysis, document summarization, multi-turn conversations. It’s not GPT-4-128k territory, but it covers 95% of real use cases.
Pros & Cons
Pros:
- Widest model support among mid-tier relays: Claude, GPT, Gemini, DeepSeek, Kimi, Grok, GLM, Qwen
- Dual pricing (pay-per-call + pay-per-token) lets you optimize cost per use case
- ¥1 minimum recharge — lowest barrier for testing
- Daily check-in rewards add value for regular users Cons:
- Cheapest channel dilutes quality — you may not get the model you paid for
- Day-to-day quality variation on some channels
- Support response time is inconsistent
- 96% uptime isn’t production-grade
- No refund policy specified — risk if you preload ¥100+
Verdict
LinkAPI is a tier-2 relay — good for exploration, prototyping, and low-stakes personal projects. The model breadth is genuinely useful, especially if you want Claude and GPT under one account without managing multiple keys. ¥1 minimum recharge makes it the cheapest entry point in the China-to-LLM relay space. Don’t use it for production workloads that require deterministic model behavior or 99.5%+ uptime. The quality dilution on cheap channels and variable daily performance are real risks. Do use it for testing multiple models side-by-side, personal coding assistants, or as a secondary relay when your primary provider has issues. The daily check-in rewards offset some cost for frequent users. If you need reliability, pair LinkAPI with a higher-tier relay as fallback. If you just want to poke at GPT-4o and Claude from China without VPN, ¥1 gets you started.
FAQ
Q: Can I use LinkAPI without a VPN in China? A: Yes. That’s the core feature — all models are accessible from mainland China through their relay infrastructure. No VPN required. Q: What happens if I select the cheapest channel and get a diluted model? A: You’ll still receive a response, but it may come from a lower-quality variant of the requested model. For critical work, explicitly choose the pay-per-token premium channel instead of the cheapest pay-per-call option. Q: How does the daily check-in reward work? A: The platform offers daily check-in bonuses that add credits to your account. Exact amounts aren’t specified in the data, but it’s a recurring discount for consistent users. Q: Is there a refund if I recharge and don’t use the credits? A: Refund policy is not specified. Given the ¥1 minimum recharge, this is low risk — only load what you’re willing to use within a few weeks. Q: How does 96% uptime compare to other relays? A: It’s below average for the relay space. Top-tier providers target 99.5-99.9%. The 14 hours/month of potential downtime makes LinkAPI unsuitable for production systems without a backup relay.
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.
