I Built a Unified AI API With 6 Models — What I Learned From 6 Months of Running It

#ai #webdev #opensource #api

Why I Built This

6 months ago, I was managing 5 separate AI API accounts. OpenAI for chat, Anthropic for code, DeepSeek for cost-sensitive tasks, Google for vision. Each with its own SDK, its own billing, its own dashboard.

It was a mess.

So I built Dubhe AI — a single API gateway that routes requests to the best model for each task, all under one account, one API key, one bill.

What's Under the Hood

Fast (DeepSeek V4 Flash) — general-purpose, great for chat
Vision (MiMo V2.5) — image understanding
Code (GLM 4.7) — programming tasks
Agent (MiniMax M2.7) — complex agent workflows
Plus (Qwen 3.6) — balanced performance
Reasoning (DeepSeek V4 Pro) — complex reasoning

All OpenAI-compatible. Switching from your current provider takes about 5 minutes.

Pricing That Actually Makes Sense

Plan	Price	Tokens/Month
Starter	9 dollars	3M
Pro	29 dollars	12M
Business	49 dollars	20M
Pro Plus	99 dollars	40M

All plans include access to all 6 models. No hidden fees, no per-model surcharges.

What I Learned

1. Price matters more than brand. Most indie devs I talked to choose providers based on cost first, performance second. Being 5-10x cheaper than OpenAI/Anthropic is a real advantage.

2. One API key is a surprisingly big deal. Developers hate managing multiple credentials. The "just works" factor is real.

3. Global payments are hard. PayPal rejected our application 3 times. We eventually went with Paddle as our merchant of record — they handle VAT, sales tax, and payment methods in 100+ countries.