Moonshot AI Releases Kimi K2.5 — 1T Parameter Open-Source Multimodal Agent Model
Published 2026-01-27Foundation ModelsMedium
Summary
Alibaba-backed Chinese AI startup Moonshot AI released Kimi K2.5, a 1 trillion-parameter open-source multimodal model using a mixture-of-experts (MoE) architecture, trained on 15 trillion mixed visual and text tokens. The model natively understands text, image, and video, activating only the specific expert networks relevant to each prompt rather than all parameters. Key differentiators: Kimi K2.5 introduces "Agent Swarm" technology, enabling the model to decompose complex tasks into simpler su
Alignment: New signal not yet covered
Related Positions: agentic-workflows.md
moonshotkimik2-5open-sourcemultimodalmixture-of-expertsagent-swarmchinafoundation-modelscoding-agentdeepseekalibaba