Moonshot AI Launches Kimi K2.6 With 300-Agent Swarm Capability, Advancing Autonomous AI Systems

Gate News message, April 21 — Moonshot AI released Kimi K2.6, an open-source AI model featuring a significant upgrade to its agent swarm architecture, now supporting up to 300 parallel sub-agents—nearly three times the 100 sub-agents supported by its predecessor K2.5. The model is accessible via Kimi.com, the Kimi app, the API, and Kimi Code, targeting developers building agent-based workflows and automation tools.

The K2.6 release scales from K2.5's 1,500 coordinated steps to 4,000, enabling greater parallelization and faster execution for complex multi-domain tasks. According to Moonshot AI, the model demonstrates improved performance in long-horizon coding across Rust, Go, and Python, with enhanced capabilities in front-end development, DevOps operations, and performance optimization. Benchmark results show state-of-the-art performance on HLE with tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), and Math Vision with Python (93.2).

The model introduces a "Skills" mechanism that converts structured documents—including PDFs, spreadsheets, and Word files—into reusable task templates, capturing formatting and reasoning patterns for future workflows. K2.6 is positioned for autonomous use cases including multi-step software workflows, full-stack application development, and persistent monitoring tasks, with testing showing improved reliability over extended execution cycles.