Hey IH — I want to share what I've been building for the past few months.
THE PROBLEM
I'm a system integrator based in Gansu, China. My company resells cloud services to education, healthcare, and government clients across northwest China. Every time a client needed an AI feature — chatbot, OCR, translation — I had to sign up with a different provider, deal with their billing, handle their SDK, and pray the API didn't break.
The Chinese AI API landscape is fragmented: Baidu, Alibaba, Tencent, ByteDance, Moonshot, DeepSeek, Zhipu — each with different auth, different pricing, different rate limits. For a 6-person team, integrating 7 providers meant 7x the maintenance.
WHAT I BUILT
GoldBean API — one endpoint, one API key, 67 endpoints across 7 Chinese LLM providers. Developers get:
TECH STACK
PRICING MODEL
Three tiers: Free ($0, 2,000 calls/day), Pro ($19.9/mo), Business ($89/mo). Token-based pricing for LLM endpoints ranges from $0.05/1M tokens (budget models) to $2.50/1M tokens (premium).
CURRENT STATUS
WHAT I LEARNED
WHAT'S NEXT
If you're building anything with Chinese AI models, I'd love your feedback. The playground is at goldbean-api.xyz/playground — you can test all 7 models side by side without signing up.
The interesting part seems less like the number of endpoints and more like whether developers actually want one abstraction layer across these providers badly enough to switch from integrating them directly. That feels like the key assumption to validate.