The 30-second summary
+ What we liked
- Aggregator that forwards from cheapest relay sources
- China latency-optimized
- ¥1 free credit on signup
- Can issue fapiao (发票) invoices
− What we didn't
- Not a direct relay — adds another hop
- Quality depends on upstream sources
- Less transparent about origin channels
In-depth review
GPT-4o input costs $0 per million tokens on MNAPI — if you stay within the ¥1 free credit.
That’s the hook. MNAPI isn’t selling direct API access. It’s an aggregator of aggregators: it forwards your requests to the cheapest upstream relay sources, takes a cut, and hands you a single API key. The result is a pricing model that undercuts almost every direct provider, but comes with tradeoffs you need to understand before routing production traffic through it.
Pricing Breakdown
MNAPI operates on a credit system. You get ¥1 free on signup — roughly $0.14 USD at current rates. After that, you recharge via 支付宝 or 微信支付 (no international card support). There’s no published minimum recharge amount, and no refund policy is specified.
The actual per-token rates aren’t listed publicly. Instead, MNAPI dynamically prices based on the cheapest available upstream relay at the moment of your request. This means:
- GPT-4o: Typically $0.50–$0.80 per million input tokens (vs $2.50 direct from OpenAI)
- Claude 3.5 Sonnet: Usually $1.00–$1.50 per million input tokens (vs $3.00 direct from Anthropic)
These are estimates based on typical relay pricing. The actual rate fluctuates. You won’t know the exact cost until you see the charge on your credit balance.
Cost Comparison vs Direct API
| Model | Direct API (per 1M input tokens) | MNAPI (estimated) | Savings |
|---|---|---|---|
| GPT-4o | $2.50 | $0.50–$0.80 | 68–80% |
| Claude 3.5 Sonnet | $3.00 | $1.00–$1.50 | 50–67% |
The savings are real. But you’re paying for them in reliability.
The Hidden Cost: Reliability
94% uptime. That’s what MNAPI reports. For a relay service, that’s below average. Compare to direct API uptime (99.9%+ for OpenAI and Anthropic) or even top-tier relays like Helpaio (99.5%+).
That 5.6% downtime gap means roughly 40 hours of unavailability per month. If your app needs consistent access to LLMs, that’s a problem.
The other hidden cost is latency. MNAPI claims “China latency-optimized,” which is true — requests route through domestic servers before hitting upstream relays. But each relay hop adds 50–200ms. Your total latency will be higher than a direct connection from a US-based server.
Batch Discounts and Free Tier
No batch discounts are advertised. The ¥1 free credit is the only promotional offer. For a developer testing an app, ¥1 covers roughly 1.25 million GPT-4o tokens or 0.7 million Claude 3.5 Sonnet tokens — enough for serious prototyping.
When MNAPI Makes Sense
You’re a Chinese developer who:
- Needs cheap access to GPT-4o and Claude 3.5 Sonnet
- Can tolerate occasional downtime
- Doesn’t need transparency about which upstream relay serves your request
- Wants fapiao (发票) invoices for business expenses
When to Skip It
You’re building a production app where:
- Uptime matters (94% won’t cut it)
- You need consistent, predictable pricing
- You want to know exactly which model version you’re hitting
- You’re outside China (the latency optimization doesn’t help you)
Pros & Cons
Pros
- Cheapest GPT-4o and Claude 3.5 Sonnet access available
- ¥1 free credit for testing
- Domestic China payment methods (支付宝, 微信支付)
- Can issue fapiao invoices for business users
Cons
- 94% uptime is unreliable for production
- No transparency on upstream sources or model versions
- Adds latency from the extra relay hop
- No refund policy specified
- Only two models available
Verdict
MNAPI is a budget option for Chinese developers who need cheap LLM access and can tolerate downtime. The ¥1 free credit is generous enough to evaluate the service. But the 94% uptime and lack of transparency make it unsuitable for production applications. If you’re prototyping or running non-critical tasks, the price is hard to beat. For anything that needs to stay online, look elsewhere.
FAQ
Q: Does MNAPI support GPT-4 and Claude 3 Opus? A: No. MNAPI only supports GPT-4o and Claude 3.5 Sonnet. Check the Models field — those are the only two available.
Q: Can I use MNAPI with a US-based server? A: You can, but the latency optimization is designed for China. US-based users will see higher latency than direct API access.
Q: How do I recharge my account? A: Via 支付宝 or 微信支付. No international credit cards or PayPal. Minimum recharge amount is not specified.
Q: What happens if an upstream relay goes down? A: MNAPI should route to another upstream, but the 94% uptime indicates this doesn’t always work smoothly. You’ll experience downtime.
Q: Can I get a refund for unused credit? A: No refund policy is specified. Assume all credits are non-refundable.
Pricing breakdown
MNAPI offers competitive pricing for developers. Here's the breakdown:
| Plan | Price | Quota | Best for |
|---|---|---|---|
| Free | $0/mo | Free trial | Kicking the tires |
| Standard RECOMMENDED | Pay-as-you-go/mo | Unlimited usage | Solo devs · small teams |
| Enterprise | Custom | SLA · dedicated support | Teams & agencies |
Supported models
2 models across major vendors.
Frequently asked questions
Can I access this platform from China without a VPN?
Most relay stations are accessible from Chinese ISPs. Check our review for specific routing details.
What payment methods are accepted?
Payment options vary by platform. Some accept Alipay/WeChat Pay, others are USD/crypto only.
How does this compare to using OpenAI directly?
Relay stations add routing latency but provide access from restricted regions, unified billing, and multi-model fallback.
Is my API key safe?
Keys are encrypted at rest. Most platforms support per-project scoping and IP allow-lists.
Should you use MNAPI?
Users who want one API key to rule them all