Moonshot Drops Kimi K2.6 — Four Models, 300 Parallel Agents
Moonshot AI upgrades all Kimi models to K2.6, adding an Agent Swarm tier that coordinates up to 300 sub-agents simultaneously.
Moonshot AI just pushed Kimi K2.6 across Kimi Chat and its API — four tiers at once: Instant, Thinking, Agent, and Agent Swarm. No waitlist, no phased rollout — live now.
The standout is Agent Swarm: up to 300 parallel sub-agents executing up to 4,000 coordinated steps, tripling K2.5's 100-agent cap. That's not a spec-sheet flex — it fundamentally changes throughput for large codebase refactors and batch generation tasks. The MoE backbone stays the same: 1.04T total parameters, 32B active per token.
K2.5 already scored 76.8% on SWE-Bench Verified and undercut Claude Opus 4.5 by ~76% on cost. K2.6 official benchmarks aren't published yet, but Moonshot's cadence — K2 (July '25), K2.5 (Jan '26), K2.6 (now) — makes clear they're not pacing themselves.
Why it matters: the Agent Swarm tier positions Kimi as a serious infrastructure-layer play for agentic coding, not just another chat model.
Sources
Primary: the company, paper or repository
- moonshotai/Kimi-K2.6 on Hugging Face huggingface.co
Independent coverage
- Moonshot AI Launches Kimi K2.6 with Enhanced Features phemex.com
- Kimi K2.6 Just Dropped — Beats Claude Opus 4.5 at 76% Lower Cost medium.com
Written by an AI pipeline from the sources above. Methodology · Report an error
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.