There’s something strange happening with AI right now.
A small model can surprise you with how much it knows. A much bigger model can spend a ridiculous amount of time doing something simple. Then you give both models one difficult problem, and suddenly the difference becomes enormous.
The interesting part is that you don’t always know which one will do better until the work begins.
I have been looking closely at this, and I think there’s a much smarter way to use all these powerful models. It’s one of those things that seems small at first, but can change how you build and use AI completely.
Your AI might be spending money for no reason
Look at the latest prices.
OpenAI’s GPT-6 Luna costs $0.10 per million input tokens. GPT-6 Astra costs $10. That’s 100 times more for the same amount of input. And both are available through the same API.
Claude has a similar range, from Haiku to Sonnet and Opus. Google, DeepSeek, Kimi, and other AI companies offer their own models with different strengths and prices.
Now imagine you’re building a small game with Claude Code. Your agent needs to find a file, change how a character moves, fix a difficult bug, and check whether the game still works.
Finding the file is easy. Fixing the bug might require serious reasoning. Checking the result is another kind of work.
Why should all three jobs use the same model?
This is where AI Model Routing comes in.
You give an AI system access to several models, and a router decides which one should handle each piece of work. Smaller models can take care of easier parts. Stronger ones can step in when the problem gets difficult.
You can even let the router make that decision automatically.
The exciting part is that this already works with tools such as Claude Code, OpenRouter, n8n, and AI agents. You don’t have to build a complicated system to get started.
Let’s make one below..
Read more
