Moonshot AI Launches Kimi K2.6 Swarm With 300 Sub-Agents
Moonshot AI introduced Kimi K2.6 Agent Swarm, saying the system can run 300 parallel sub-agents with 4,000 steps per run, up from 100 sub-agents and 1,500 steps in K2.5. The company said the model produces files rather than chat-only responses, with a single run able to generate more than 100 files, a 100,000-word literature review or a 20,000-row dataset; one example showed 40 PDFs turned into a 100,000-word literature review and cited dataset.
Nous Research has opened free access to the newly launched Kimi K2.6 on the Nous Portal for 24 hours through Vercel's AI Gateway. OpenRouter's programming leaderboard lists Kimi K2.6 at No. 1, while Arena placed it as the No. 2 open model in Code Arena and No. 6 overall.
From the sources (7 posts)
@alexatallahRT @Kimi_Moonshot: Kimi K2.6 is now ranked #1 on OpenRouter's programming leaderboard.
@wesrothNous Research has temporarily opened free access to Moonshot AI's newly launched "Kimi K2.6" model on the Nous Portal for 24 hours, facilitated by Vercel’s AI Gateway.
@wesrothso Kimmy k2.6 is free right now in Hermes Agent... been running it all day... it's *surprisingly* good... open source is really closing the gap ngl, this feels like like the first open source model worth using for serious projects @NousR
@kimi_moonshotMeet Kimi K2.6 Agent Swarm 👋 Highlights： 🔹 Swarms, elevated - 300 parallel sub-agents × 4,000 steps per run (up from 100 / 1,500 in K2.5). 🔹 Outputs are real files, not chat - one run delivers 100+ files, 100,000-word literature reviews
@kimi_moonshotWhat it ships in one run: 🔹 100 job description → 100 tailored CVs
@kimi_moonshot🔹 40 PDFs → 100,000-word literature review + cited dataset
@nandodfRT @arena: Kimi-K2.6 is now live in the Arena, and it’s a big improvement over Kimi-K2.5: - #2 open model in Code Arena (#6 overall), on pa…