Hey IH! I built an OpenAI-compatible API that costs 90% less than GPT-4.
The problem: DeepSeek is incredibly cheap ($0.14/M tokens vs GPT-4's $2.50/M), but signing up requires a Chinese phone number. Many developers outside China can't use it.
So I built a proxy: https://api.aifreeaistack.com
- ✅ Drop-in OpenAI replacement (just change the base_url)
- ✅ No Chinese phone number needed
- ✅ Pay as you go (starting at $10)
- ✅ DeepSeek Chat + DeepSeek Reasoner
- ✅ Works with any OpenAI SDK (Python, Node.js, curl)
Pricing: ~$0.69/M tokens input — still 3.6x cheaper than GPT-4o.
Built with Flask + SQLite. Took about a week.
Would love feedback from the community! What's missing?
I like that you're removing an adoption barrier rather than trying to build another LLM.
The interesting challenge is making developers see the product as a reliability layer, not just a cheaper endpoint. Cost gets people to try it, but reducing the friction of accessing and integrating models is what keeps them around.