Show HN: RouteGPT – model routing on ChatGPT aligned to user preferences
I had a personal itch to scratch, and thought that it might be useful to others: If you are a ChatGPT pro user like me, you are probably tired of pedaling to the model selector drop down to pick a model, prompt that model and then repeat that cycle all over again. So I built RouteGPT to end that pedaling.
RouteGPT is a Chrome extension for chatgpt.com that automatically selects the right OpenAI model for your prompt based on preferences you define. For example: “creative novel writing, story ideas, imaginative prose” → GPT-4o, or “critical analysis, deep insights, and market research ” → o3
How does it work?
Underneath the covers, RouteGPT decouples the routing decision into two phases: route selection and model assignment.
Route Selection: This is the what. You define a set of human-readable routing policies using a “Domain-Action Taxonomy.” Think of it as clear usage scenarios that can be captured in plain English. More specifically, you define routing policies using a Domain-Action Taxonomy (e.g., healthcare, code explanation) expressed in natural language. The route selection is predicted by a small (yet powerful) 1.5b LLM router model [1]. You can read more about the research in the paper [2]
Model Assignment: This is the how. A separate, simple mapping configuration connects each policy to a specific LLM. The finance/analyze_earnings_report policy might map to a powerful model like GPT-4o, while a simpler general/greeting policy maps to a faster, cheaper model.
Hope you all enjoy the extension - was fun "vibe coding" it up.
[1] Model: https://huggingface.co/katanemo/Arch-Router-1.5B [2] Paper: https://arxiv.org/abs/2506.16655
P.S Btw if you want to use this type of preference-aligned routing for your chatbots, its fully packaged and integrated as part of Arch: the open source edge and service proxy that I am built: https://github.com/katanemo/archgw
No comments yet