Moonshot AI has released Kimi K2.6 as an open-weight model. It's built to match GPT-5.4 and Claude Opus 4.6 on coding benchmarks, and it can run up to 300 agents in parallel.
Moonshot AI says K2.6 puts up top scores across several benchmarks, landing on par with GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro. The numbers include 54.0 on HLE with Tools, 58.6 on SWE-Bench Pro, and 83.2 on BrowseComp. The model can chain together more than 4,000 tool calls and run continuously for over twelve hours in languages like Rust, Go, and Python.
Kimi K2.6 keeps pace with the top models from OpenAI, Anthropic, and Google on coding and agent benchmarks, though it falls behind on pure reasoning and vision. | Image: Kimi
300 agents working in parallel
The headline feature is Agent Swarm, which can run up to 300 sub-agents at once, each taking 4,000 steps. The system automatically splits tasks into subtasks and hands them off to specialized agents. Moonshot AI says these agents combine skills like web research, document analysis, and writing, and a single run is meant to produce finished outputs, including documents, websites, slide decks, and spreadsheets. Here's an example:
