Kimi K2.6: Open-Weight 1T MoE Beats GPT-5.4 on SWE-Bench Pro (58.6 vs 57.7) with 300-Subagent Swarm Scaling at $0.60/M Input

MarkTechPost·medium signal

Moonshot AI released Kimi K2.6 on April 20 — a 1 trillion parameter open-weight MoE model scoring 58.6 on SWE-Bench Pro (vs GPT-5.4's 57.7 and Claude Opus 4.6's 53.4). The model supports agent swarm scaling to 300 sub-agents and 4,000 coordinated steps for long-horizon coding tasks. Kimi Code CLI (6,400+ GitHub stars) now ships with K2.6 as default backend at $0.60 input / $2.50 output per million tokens under Modified MIT License.