Cloudflare AI Gateway unifies model access

This is the kind of infrastructure update that looks minor until it becomes the default abstraction for teams juggling multiple model vendors. The real value is less about syntax and more about making provider churn and model churn mostly invisible to application code.

• –`env.AI.run()` now accepts both `@cf/...` Workers AI models and third-party `{author}/{model}` models through AI Gateway
• –Cloudflare’s docs position AI Gateway as the control plane for unified billing, caching, logs, and provider routing
• –Third-party models are supported across a broad provider set, so teams can standardize on one integration surface instead of hard-coding vendor APIs
• –BYOK is not supported through the binding for third-party models, which pushes some advanced users toward the REST or OpenAI-compatible endpoints instead
• –Inference: Cloudflare is trying to own the “AI routing layer” for apps, not just host its own models