Moonshot AI Releases Kimi K2.6 Open-Source Model with Advanced Agent Workflows

Gate News message, April 22 — Moonshot AI released its open-source Kimi K2.6 model on April 20, introducing new capabilities for coding, front-end generation, and multi-agent orchestration. The model is designed to support complex enterprise workflows through advanced Agent Programming and Deep Research and Reasoning features.

Kimi K2.6 is built for extended coding projects and coordination across multiple AI agents. The earlier K2.5 model's Agent Swarm feature can launch up to 100 sub-agents in parallel, and the new version supports up to 300-step tool calling for layered workflows in competitive analysis and complex legal document processing. On some benchmarks, Moonshot said Kimi K2.6 matched or exceeded closed-source models including GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro, though the company noted evaluation standards remain uneven and results lack independent verification.

Moonshot is expanding beyond model releases by building an open-source developer ecosystem, maintaining an active GitHub presence with tools such as a command-line interface (CLI) agent. The model has gained integration into external platforms, including Vercel's AI Software Development Kit. Industry-specific applications using Kimi include a FinGPT Agent for financial analysis, plus tools for legal, document, and pharmaceutical research workflows. The release reflects a broader trend among Chinese AI firms—including Alibaba, ByteDance, and Tencent—adopting mixed strategies with both open-source and closed-source model offerings.