Cloudflare Adds Kimi K2.5, Cuts AI Agent Security Audit Costs by 77%

Cloudflare’s Workers AI is a serverless GPU inference platform. As the use of coding and personal AI agents grows rapidly, the cost of long-context inference has become a barrier to scaling. Cloudflare therefore selected Moonshot AI’s open-source Kimi K2.5 model, which combines a 256K context window and multi-turn tool calling with cost efficiency.

Cloudflare announced the addition of Kimi K2.5 to Workers AI on March 19, 2026. Its security audit agent processes more than 7 billion tokens a day. The switch cut costs by 77%, saving about $1.85 million annually compared with the roughly $2.4 million yearly cost of a mid-tier proprietary model. Cloudflare also added prefix caching, session affinity and an asynchronous API to improve efficiency in long conversations.