Powering the agents: Workers AI now runs large models, starting with Kimi K2.5

Cloudflare has announced a significant development in its AI capabilities, as it introduces the Kimi K2.5 model to its Workers AI platform. This move marks a major step forward in enabling developers to build and deploy agents directly on the Cloudflare Developer Platform, providing a unified infrastructure for the entire agent lifecycle.

The Cloudflare Developer Platform has long been a robust environment for building and executing applications, thanks to its suite of tools such as Durable Objects for state management, Workflows for long-running tasks, and Dynamic Workers or Sandbox containers for secure execution. However, these primitives primarily offered an execution environment, leaving the need for a powerful AI model to power the agents.

To address this gap, Cloudflare has now entered the big models game by offering frontier open-source models on its AI inference platform. The company is starting with Moonshot AI's Kimi K2.5 model, which boasts a full 256k context window and supports multi-turn tool calling, vision inputs, and structured outputs. This model is particularly well-suited for various agentic tasks, offering a high degree of reasoning capabilities and efficiency.

By integrating Kimi K2.5 directly into the Cloudflare Developer Platform, the company is making it possible to run the entire agent lifecycle on a single, unified platform. This integration not only simplifies the development process but also ensures that the AI model powering the agent is optimized for performance and cost-efficiency.

Cloudflare has already tested Kimi K2.5 as the engine for its internal development tools, particularly within the OpenCode environment. Cloudflare engineers have been using Kimi as a daily driver for agentic coding tasks, and the model has been integrated into the company's automated code review pipeline. This integration is evident in the public code review agent, Bonk, which operates on Cloudflare GitHub repositories.

In production, Kimi K2.5 has proven to be a fast and efficient alternative to larger proprietary models without compromising on quality. Initially launched as an experiment, the model quickly became critical to Cloudflare's internal operations, demonstrating its value and potential for broader adoption.

This development underscores Cloudflare's commitment to making its platform the best choice for building and deploying agents. By offering a robust infrastructure and integrating cutting-edge AI models like Kimi K2.5, the company is positioning itself at the forefront of the agent development ecosystem. As Cloudflare continues to expand its AI offerings, developers can expect even more powerful and efficient tools to support the creation of intelligent, agentic applications.