Neuralrouting.io

Hey IH 👋

I’m building a product in the AI cost optimization space and I want to reality-check my assumptions before going further.

If you’re running an app that makes API calls to OpenAI, Anthropic, Groq, etc., I’d love to hear:

1. How much are you spending per month on LLM APIs? (ballpark is fine — $50? $500? $5,000?)

2. What percentage of your prompts actually need GPT-4o / Claude Sonnet? In my experience building AI apps, ~70% of requests are simple enough for a cheaper model.

3. Have you tried routing between models to save money? If yes, what worked? If no, what stopped you?

4. Would you pay $29/mo for a proxy that automatically picks the cheapest model that can handle each request? Drop-in replacement — just change your base_url, no other code changes.

Some context: I built NeuralRouting.io — an intelligent API gateway that analyzes each prompt in real-time and routes it to the optimal model. Simple questions go to Llama (nearly free), complex ones go to GPT-4o. It also includes a quality monitoring engine that catches when the cheap model gives a bad answer and auto-escalates.

But honestly, I’m more interested in hearing about YOUR experience with LLM costs than pitching. The product is live and working, but I want to make sure I’m solving a problem people actually have — not one I imagined.

Drop a comment or DM me. Happy to share what I’ve learned about model routing if it’s useful.