Moonshot releases Kimi K2.5, a multimodal agent model

Moonshot AI/Kimi models

Moonshot released Kimi K2.5, an open-weight 1-trillion-parameter model with 32 billion active parameters that understands images and video natively and can split big tasks among a swarm of sub-agents working in parallel. It is especially aimed at writing code from visual designs, with a 256K-token context, under a modified MIT licence.

• 1-trillion-parameter open-weight model with 32 billion active per token; 256K-token context under modified MIT licence
• Native multimodal understanding of images and video enabling code generation from visual specifications and autonomous tool orchestration
• Decomposes complex tasks into parallel sub-tasks executed by dynamically created specialised agents
• Offers thinking and instant modes with interleaved reasoning; competes with GPT-5.2 and Claude 4.5 Opus on various benchmarks