Kimi K2.5 Achieves Large-Scale Multimodal Agentic Capabilities
Moonshot AI releases Kimi K2.5, a 1-trillion-parameter visual agentic intelligence model with 32 billion active parameters, joint vision-text pretraining on roughly 15 trillion tokens, and post-trained checkpoints under a Modified MIT license. The release emphasizes multimodal RL, Agent Swarm and PARL for up to 100 subagents and 1,500 tool calls, plus inference optimizations yielding up to 4.5× latency reduction, implying practical multimodal agents for complex, parallel workflows.
Scoring Rationale
Breakthrough model scale, multimodal RL and public checkpoints drive high impact; strength in architecture, limited independent replication so far.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
