The 30-second summary
+ What we liked
- MAX group — no dilution on Claude models
- Supports pay-per-call pricing (¥0.05/call)
- Codex available with GPT-5.5
- ¥1 minimum recharge
− What we didn't
- New station (March 2026) — limited track record
- Relatively high pricing on MAX groups
- Limited model selection
In-depth review
GPT-5.5 costs ¥0.05 per call via 发现AI, but there’s no per-token pricing disclosed — you pay per API call, not per million tokens.
This is the first thing to understand about 发现AI’s pricing model: it’s a pay-per-call system, not the per-token billing you’re used to from OpenAI or Anthropic. At ¥0.05 per call, that’s roughly 7 US cents at current exchange rates. For a single short completion, that’s cheaper than GPT-4o’s direct API. For a long 4K-token response, it’s more expensive. There’s no free tier beyond the initial trial — the “free_trial: True” flag suggests new users get some test credits, but the platform data doesn’t specify how many.
Pricing Structure
| Model | Price per call | Min recharge | Payment methods |
|---|---|---|---|
| GPT-5.5 | ¥0.05 | ¥1 | 支付宝, 微信支付 |
| Claude Opus 4.6 | ¥0.05 (MAX group) | ¥1 | 支付宝, 微信支付 |
| Codex | Included with GPT-5.5 | ¥1 | 支付宝, 微信支付 |
The ¥1 minimum recharge is aggressively low — you can test with less than 15 US cents. That’s rare among Chinese relay stations, which typically require ¥10-50 minimums. The catch: you’re buying calls, not tokens. If your workflow sends 100 short prompts per day, that’s ¥5/day or ¥150/month. For heavy users generating 4K+ token responses, direct API access through a VPN may be cheaper per million tokens.
Cost Comparison vs Direct API
Direct GPT-4o pricing (2026): roughly $2.50 per million input tokens, $10 per million output tokens. At ¥0.05/call, 发现AI breaks even around 1,000 output tokens per call. Shorter calls save you money. Longer calls cost more. There are no hidden fees — the pricing is transparently per-call. No batch discounts exist in the platform data.
The “MAX group — no dilution on Claude models” pro means Claude Opus 4.6 calls aren’t shared across users. You get dedicated capacity, which explains the ¥0.05 flat rate. Competing stations often pool Claude requests, causing 5-15 second latency spikes during peak hours. 发现AI avoids that by charging a premium.
Pros & Cons
Pros
- ¥1 minimum recharge — lowest barrier I’ve seen for Chinese relay stations
- Pay-per-call model predictable for short prompts (no surprise token bills)
- Claude Opus 4.6 has dedicated capacity (MAX group)
- 支付宝 and 微信支付 — no need for international credit cards
Cons
- No per-token pricing — hard to compare cost efficiency for long outputs
- Only 3 models available (GPT-5.5, Claude Opus 4.6, Codex)
- 94% uptime is below the 98-99% standard for established stations
- New station (March 2026) — limited reliability data
- No batch discounts or volume pricing
Verdict
发现AI works for one specific use case: developers who need short, occasional API calls to GPT-5.5 or Claude Opus 4.6 without VPN, and want to spend less than ¥10 total. The ¥1 minimum recharge makes it a zero-risk test. Don’t use this for production workloads with long outputs or high throughput — the per-call model gets expensive fast, and 94% uptime means you’ll hit downtime roughly 2 hours per month. For heavy usage, look at platforms that publish per-million-token rates and offer batch discounts.
FAQ
Q: How does ¥0.05/call compare to GPT-4o’s per-token pricing? A: For responses under ~1,000 output tokens, 发现AI is cheaper. Above that, direct API via VPN is more cost-effective. There’s no way to know exact breakpoints without per-token rates from 发现AI.
Q: Is there a free tier for testing? A: The platform data indicates a free trial exists, but doesn’t specify how many free calls or credits you get. Expect a small allocation — likely 5-10 calls — given the ¥1 minimum recharge.
Q: Can I use Codex without GPT-5.5? A: No. Codex is listed as “available with GPT-5.5,” meaning it’s bundled. You can’t purchase Codex calls independently.
Q: What happens if I exceed ¥1 in calls? A: The platform doesn’t specify auto-top-up behavior. With ¥1 minimum recharge, you likely need to manually add funds when your balance runs out. No auto-recharge details are available.
Q: Does 发现AI support streaming responses? A: Not mentioned in the platform data. Given the pay-per-call model and limited model selection, streaming is unlikely. Expect standard request-response latency.
Pricing breakdown
发现AI 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
3 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 发现AI?
New users wanting to test with minimal commitment