Cloudflare merges Workers AI and AI Gateway into one control plane
Cloudflare is unifying Workers AI and AI Gateway into a single control plane, bringing automatic observability, unified billing, and a path to model-first routing.
Cloudflare is collapsing the boundary between Workers AI and AI Gateway. What started as two separate products — one for hosted model inference, one for proxying to any provider — is now converging into a single control plane. The goal: one entrypoint, one billing system, one set of logs, and eventually routing that cares about the model, not the provider.
One binding, one API, no more choice
The first step is unifying the entrypoints. The Workers AI binding and the REST API now both route through the same path, with a gateway option that points at a gateway ID. If you've never created a gateway, passing default auto-creates one on first use. The same goes for the REST API, where a cf-aig-gateway-id header does the trick.
This means you no longer have to decide between Workers AI and AI Gateway upfront. Every call gets observability, logging, and cost tracking by default, with the option to create named gateways later for custom caching or traffic splitting.
Observability without setup
Previously, getting visibility into Workers AI traffic required explicitly setting up an AI Gateway. Now, the default gateway gives you request/response logging, token counts per model, and cost attribution automatically. The dashboard shows latency breakdowns, error rates, and exact prompts and responses — a significant upgrade for teams debugging model behavior or auditing outputs.
Unified billing and elevated rate limits
Cloudflare is also enabling AI Gateway credits to be used for Workers AI. You can load a wallet and spend it across OpenAI, Anthropic, Workers AI, or any supported provider. As an incentive, using unified billing gets you elevated rate limits on Workers AI models, with details in the developer docs.
Model-first routing on the horizon
The bigger play is model-first routing. Instead of specifying a provider, you specify a model and let the gateway handle provider selection, failover, and load balancing. If Workers AI has capacity, you get managed infrastructure; if it's at capacity, traffic shifts to another provider hosting the same weights. This is designed to improve resiliency without application-level retries.
Beyond that, Cloudflare is working on smart routing: a classifier on Workers AI reads your prompt, predicts task type and complexity, and picks the best model from a curated pool. This is still in internal pilot, but the zero-config path could remove the need to maintain your own routing logic.
Get started
Existing Workers AI users can start routing through a default gateway to get logging and cost attribution immediately. AI Gateway users can add Workers AI by simply calling a model and loading the wallet. The unified control plane is available now, with model-first routing piloting in the coming months.
Model-first routing flips the script: you think about what you need — a capable reasoning model, a fast summarizer, a cheap embedding model — and the control plane handles provider selection, failover, and load balancing.