Moonshot AI has released Kimi K2.5 as open source, adding native agent swarm orchestration to a multimodal coding and vision model positioned as an alternative to closed systems from U.S. labs.
Moonshot AI has upgraded and open-sourced its Kimi K2 model as Moonshot Kimi K2.5, adding built-in agent swarm orchestration that allows autonomous agents to coordinate tasks without a central orchestration layer.
The model supports agent-based systems in which up to 100 specialised sub-agents can operate in parallel and execute workflows involving up to 1,500 tool calls. Instead of increasing model size, the approach relies on distributing work across multiple agents that operate concurrently, a structure the company compares to a beehive.
Kimi K2.5 is positioned as an open alternative to closed models from Google, OpenAI, and Anthropic, with benchmark results placing it in the same performance range. On Humanity’s Last Exam, it scored 50.2% with tools, exceeding results reported for OpenAI GPT-5.2 (xhigh) and Claude Opus 4.5. On SWE-bench Verified, it achieved 76.8%, indicating strong coding performance, though trailing the highest scores reported for GPT-5.2 and Opus 4.5.
Moonshot describes Kimi K2.5 as an “all-in-one model” that accepts text, image, and video inputs. It supports visual coding workflows such as reconstructing website code by analysing screen recordings and includes visual debugging capabilities through Kimi Code, a terminal-based tool that integrates with VSCode and Cursor.
“Benchmarks only tell half the story. Moonshot AI believes AGI should ultimately be evaluated by its ability to complete real-world tasks efficiently under real-world time constraints,” the company said. “Running in parallel substantially reduces the time needed for a complex task — tasks that required days of work now can be accomplished in minutes.”
Kimi K2.5 is released under a Modified MIT License, which permits use, modification, and commercial deployment while requiring attribution for hyperscale commercial services. The company has also reduced API pricing and reports a sharp increase in user adoption following recent Kimi releases.
