Cloudflare Releases Auto Router for AI Model Selection

The new tool automates model assignment to balance performance and costs for developers using the AI Gateway platform.

Updated on Oct. 26, 2026 in Artificial Intelligence

Bold flat-color editorial illustration showing a stylized modular network hardware stack, representing AI task routing infrastructure. — Cloudflare launched its Auto Router tool in public beta, an AI Gateway feature that automatically assigns tasks to optimal models based on performance and cost. AI Illustration. Upload story photo >

Live Poll

Do you trust automated routing systems to correctly handle high-stakes business decisions?

Cloudflare has launched its Auto Router in public beta, a new feature within its AI Gateway that automatically assigns AI tasks to the most suitable models. The system evaluates incoming requests in real time to balance output quality against operating costs.

Why it matters

By automating model selection, this tool enables developers to optimize the trade-off between model performance and expenses in large-scale AI applications. It serves as an infrastructure layer for managing complex multi-model workflows efficiently.

The system processes request classification using a text cap of 8 KiB. It utilizes a scoring formula where utility equals expected quality minus an adaptive cost penalty, with requests rated on a 1 to 5 scale across 14 task categories.

The players

Cloudflare

A global provider of edge computing and cloud security services that manages massive volumes of web traffic and developer infrastructure.

The details

The Auto Router functions as a classification engine hosted on edge GPUs—specialized processors designed for parallel computation at the network edge. When a request enters the AI Gateway, a multi-head classification model analyzes the text to determine the task category and complexity. The system then filters a candidate pool of models, assigning weights by hand to the scoring matrix to optimize for both output strength and cost.

October 26, 2026: Cloudflare published the release details.

The Tech Race

This development follows an industry trend of moving model-switching logic from the application level to the infrastructure provider. Cloudflare's roadmap includes a planned auto-best feature aimed at selecting for maximum quality with no cost constraints, signaling a shift toward abstracting model management for developers.

Developers can now access the Auto Router in public beta at no additional cost while using the AI Gateway service. The feature is currently limited to non-WebSocket requests and requires configuring model pools via standard headers.

The takeaway

This utility simplifies how developers balance performance budgets for AI workflows. Watch for the future release of the auto-best tier, which will aim to prioritize output quality above all other factors.