Moonshot AI launched Kimi K2.5, scoring 50.2% on Humanitys Last Exam to surpass GPT-5.2. This 1T-par...
The AMW Read
Moonshot AI updates the player map as a top-tier lab by surpassing GPT-5.2 benchmarks and validates the scaling/agentic debate via a deliberate 1T-parameter open-weight/agent-swarm strategy.
NoveltySignificance
Foundation Models ยท Player MapScaling Laws
Moonshot AI launched Kimi K2.5, scoring 50.2% on Humanitys Last Exam to surpass GPT-5.2. This 1T-parameter model uses an Agent Swarm paradigm to run 100 sub-agents in parallel, reducing task time by 4.5x through 1,500 tool calls. The shift toward parallel execution marks a critical milestone where open-source models now lead in complex reasoning and autonomous debugging. We are entering an era where agentic coordination defines the frontier of AI productivity. ๐๐ป๐
