Cloudflare Unifies Workers AI and AI Gateway Access, Billing and Controls

Cloudflare combines two AI services

Cloudflare has unified Workers AI and AI Gateway, bringing model access, billing and traffic-management controls under a common interface, according to the company’s developer changelog and documentation.

The change allows developers to use Cloudflare-hosted models and supported third-party model providers through the same AI binding or HTTP API. Cloudflare documents the shared interfaces as the env.AI.run() binding and /ai/ REST endpoints, both using Cloudflare authentication.

Workers AI and AI Gateway previously addressed different parts of the AI application stack. Workers AI provides access to models running on Cloudflare-managed GPU infrastructure. AI Gateway sits in front of model providers and supplies functions including observability, logging, caching, security controls and billing management, Cloudflare says.

A single credit balance

The most immediate operational change is unified billing. Cloudflare’s Unified Billing documentation says customers can use one credit balance and receive one bill for Workers AI and supported third-party providers.

That structure could simplify deployments that use more than one model source. Instead of maintaining separate access patterns and balances for models run through Workers AI and models reached through AI Gateway, developers can use the same binding or API route, according to Cloudflare.

RuntimeWire reported that the consolidation also brings a common set of traffic controls across Cloudflare-hosted and external models. Cloudflare’s changelog specifically describes AI Gateway as providing observability, logging, caching, security and billing controls alongside the unified model-access layer.

Routing features are not yet broadly positioned as finished

Cloudflare is also developing capabilities intended to make model choice more flexible. RuntimeWire reported that model-first failover and prompt-based routing remain in testing.

That distinction matters: the shared access and billing structure is documented as available, while automated decisions about which model should receive a request are not presented in the supplied sources as a fully released general capability. Developers considering multi-provider architectures should therefore separate the currently documented control-plane changes from the routing features still being tested.

What the consolidation means

The update reflects Cloudflare’s effort to treat model providers as infrastructure that can be accessed through a consistent application interface. For teams building AI features on Cloudflare, the practical effect is a more unified path for invoking models, applying gateway controls and tracking spending.

Cloudflare’s documentation does not suggest that every provider or model is interchangeable. Unified billing applies to Workers AI and supported third-party providers, and the available sources do not provide a complete provider list or describe the conditions for individual models. Still, the release gives developers a common integration point for the services Cloudflare supports.

For Cloudflare, the move connects model execution and model gateway functions more closely. For developers, it reduces the number of interfaces involved in combining Cloudflare-hosted inference with external model access while retaining gateway-level logging, caching and security tools.