JA EN
shiichan

Kimi K2.6 lands on Workers AI on Day 0!

Cloudflare Changelog ★★★ View Markdown
#cloudflare#ai#serverless

Hey everyone, it's Shii-chan! Today a big new model just dropped onto Workers AI, and I'm so excited to tell you about it!

Cloudflare Changelog developers.cloudflare.com

What was announced?

On Cloudflare's Changelog, Moonshot AI's Kimi K2.6 is now available on Workers AI. The model ID is @cf/moonshotai/kimi-k2.6, and it landed with Day 0 support — you can call it from Cloudflare's edge the moment it launched.

Kimi K2.6 uses a Mixture-of-Experts architecture with 1T total parameters and 32B active per token. On agentic and coding benchmarks it competes with GPT-5.4 and Claude Opus 4.6, scoring 83.2 on BrowseComp, 80.2 on SWE-Bench Verified, and 66.7 on Terminal-Bench 2.0.

Why it matters

Frontier-scale models used to mean signing up for a dedicated API or hopping across services. Now you get one through a single Workers AI binding, so you can reach a cutting-edge model right from your own Worker. That's the nice part!

What changes

Kimi K2.6 is built to drive agents. Here's the quick rundown:

  • A 262.1k token context window that holds full conversation history, tool definitions, and codebases across long agent sessions
  • Long-horizon coding across Rust, Go, and Python
  • Coding-driven design that turns prompts and visual inputs into production-ready interfaces and full-stack workflows
  • Agent swarm orchestration scaling to 300 sub-agents running 4,000 coordinated steps
  • Vision inputs, a configurable thinking mode, and multi-turn tool calling

Dive Deep

You can reach Kimi K2.6 through the Workers AI binding, the REST API, an OpenAI-compatible endpoint, and via AI Gateway.

If you're migrating from Kimi K2.5, watch two API changes: reasoning is now controlled by chat_template_kwargs.thinking instead of chat_template_kwargs.enable_thinking, and reasoning content comes back in the reasoning field instead of reasoning_content.

For the full specs, see the Kimi K2.6 model page, and check the pricing page for costs.

Wrap-up

  • Moonshot AI's Kimi K2.6 arrives on Workers AI with Day 0 support (@cf/moonshotai/kimi-k2.6)
  • MoE with 1T parameters / 32B active, 262.1k token context window
  • Strong at agents and coding, with swarm orchestration, vision, and a thinking mode
  • Migrating from K2.5 means updated thinking control and reasoning output fields

If you want to drop a frontier-scale agentic model into your own Worker, this one's for you!

Related posts