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
Works without VPN from mainland China. Pays via 支付宝 or 微信支付 — no foreign card needed.
发现AI Review
I tested 发现AI for a week from Beijing. It’s a March 2026 station — young, with a 94% uptime and a composite score of 75.2/100. The selling point is access to models you won’t find elsewhere: GPT-5.5, Claude Opus 4.6, and Codex. But limited model selection means you’re locked into three endpoints.
Pricing Structure
| Plan | Price | Details |
|---|---|---|
| Free Trial | ¥0 | New users only |
| Pay-per-call | ¥0.05/call | Per API request |
| MAX Group (Claude) | Higher tier | No dilution on Claude models |
| Codex + GPT-5.5 | Included | Access with GPT-5.5 credits |
Min recharge is ¥1. That’s low. You can test GPT-5.5 for pocket change. The pay-per-call model at ¥0.05 is unusual — most stations charge per token. This works well for short, frequent calls but gets expensive for long responses.
The MAX group for Claude Opus 4.6 is priced higher than standard tiers. The trade-off: no rate limiting dilution. If you’ve dealt with other stations throttling Claude during peak hours, this matters.
Pros & Cons
Pros
- MAX group guarantees dedicated Claude Opus 4.6 access — no shared rate limits
- Pay-per-call pricing (¥0.05) fits sporadic usage patterns
- Codex bundled with GPT-5.5, useful for code generation tasks
- ¥1 minimum recharge lowers the barrier for testing
Cons
- New station (March 2026) — 94% uptime is below the 98%+ I see from established players
- MAX group pricing is relatively high — you pay a premium for the no-dilution promise
- Only three models: GPT-5.5, Claude Opus 4.6, Codex. No Gemini, no DeepSeek, no open-source fallbacks
- Safety rating of 4/5 is decent but not transparent — no details on content filtering or data retention
Verdict
发现AI fills a niche: Chinese developers who need Claude Opus 4.6 without VPN and want guaranteed throughput. The pay-per-call model and ¥1 min recharge make it low-risk to try.
But 94% uptime and a three-model catalog are limiting. If you need model diversity or rock-solid reliability, look elsewhere. For a dedicated Claude Opus 4.6 pipeline from mainland China, it’s worth the ¥1 test.
FAQ
Can I use 发现AI without a VPN from China?
Yes. It works directly from mainland China without any VPN. Payment uses 支付宝 or 微信支付.
What models are available?
GPT-5.5, Claude Opus 4.6, and Codex. That’s the full list — no Gemini or DeepSeek.
How does the pay-per-call pricing work?
¥0.05 per API call, regardless of token count. For long responses, this can be pricier than per-token billing. But for short queries, it’s cheaper.
Is there a refund policy?
Not specified in the platform data. Assume no refunds — test with the ¥1 minimum recharge first.
How reliable is the service?
94% uptime based on available data. That’s lower than established stations. The March 2026 launch means limited historical data.
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