Cloudflare’s AI Platform: an inference layer designed for agents

Cloudflare is evolving AI Gateway into a unified inference layer that lets developers access 70+ models from 12+ providers through a single API and one set of credits. Key updates include: using the same AI.run() Workers binding to call third-party models (OpenAI, Anthropic, Google, etc.) with a one-line switch, centralized cost monitoring with custom metadata breakdowns, automatic failover routing when a provider goes down, and streaming response buffering for resilient long-running agents. Cloudflare is also enabling developers to bring their own fine-tuned models to Workers AI via Replicate's Cog containerization technology. The platform now includes multimodal models (image, video, speech) and large agent-optimized models like Kimi K2.5, all served from Cloudflare's 330-city global network to minimize latency.