A secondary report dated August 7, 2026 states that Cloudflare announced the convergence of Workers AI and AI Gateway into a single AI control plane. Per the report, the Workers binding and REST API now route through one path (no distinct Workers AI vs AI Gateway binding), and Cloudflare introduced a unified /ai/ REST The same report describes a model-first routing model that automatically selects the provider for a given model name, alongside observability, logging, access control, and security features previously limited to the AI Gateway proxy. The summary is sourced from a single secondary outlet; no first-party Cloudflare blog

Cloudflare published a technical breakdown of how it streamlines large-scale inference on Workers AI, focusing on quantization to reduce GPU memory footprint for massive models. The post details INT4 quantization of Z.ai's GLM-5.2, shrinking it from 705GB (FP8) to 421GB with negligible benchmark degradation across GSM8 The piece also covers Cloudflare's deployment of Moonshot AI's Kimi series and Z.ai's GLM series at scale on Workers AI, explaining that smaller quantized weights improve decoding speed by reducing GPU memory reads during inference. It positions INT4 serving as a practical path to faster, cheaper, and safer production

Cloudflare's official Workers AI Changelog documents several recent model catalog updates, including the June 16, 2026 addition of @cf/zai-org/glm-5.2, Z.ai's flagship agentic coding model with a 262,144-token context window, function calling, and reasoning support. On June 12, 2026, Moonshot AI's Kimi K2.7 Code was ad A May 8, 2026 entry outlines planned model deprecations to make room for newer releases. On May 30, 2026, requests to @cf/moonshotai/kimi-k2.5 will be automatically aliased to @cf/moonshotai/kimi-k2.6 at a higher price, and a broad set of legacy models—including multiple Llama 2/3 8B variants, Llama 3.1 70B, Mistral 7B