Keep pulling the thread on Yang Zhilin.
Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks.
The Agent Swarm framework reduces latency by up to 4.5 times over single-agent baselines.
The post-trained Kimi K2.5 model checkpoint has been released to facilitate future research and real-world applications.
Kimi K2.5 is an open-source multimodal agentic model designed to advance general agentic intelligence.
The development of Kimi K2.5 utilizes techniques such as joint text-vision pre-training, zero-vision Supervised Fine-Tuning (SFT), and joint text-vision reinforcement learning.
Kimi K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework.
The Agent Swarm framework dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently.