Kimi Launches Kimi K2.5: Pioneering Visual Agentic Intelligence
Summary:Kimi has launched Kimi K2.5, the latest open-source model that builds on extensive vision-text joint pre-training to deliver state-of-the-art capabilities in visual coding and agentic intelligence. The model introduces a novel Agent Swarm paradigm, enabling up to 100 sub-agents to execute parallel workflows, significantly reducing task completion time. Kimi K2.5 excels in coding, visual reasoning, and front-end development, showcasing groundbreaking advancements in visual-to-code generation and
Kimi Launches Kimi K2.5: Pioneering Visual Agentic Intelligence
On September 7, 2026, Kimi officially released its latest open-source model, Kimi K2.5. Building on the foundation of Kimi K2, this model has achieved significant breakthroughs in visual coding and agentic intelligence through extensive pre-training on approximately 15 trillion mixed vision and text tokens.
Key Technical Highlights
-
Agent Swarm Technology
- Kimi K2.5 can coordinate up to 100 sub-agents to execute parallel workflows, significantly improving the efficiency of complex tasks. Compared to single-agent execution, Kimi K2.5 reduces task completion time by 4.5 times.
- This technology automatically creates and orchestrates agent swarms without predefined sub-agents or workflows.
-
Vision-Text Joint Pre-Training
- Through large-scale vision-text joint pre-training, Kimi K2.5 excels in visual coding and reasoning.
- The model can convert simple conversations into complete front-end interfaces, implementing interactive layouts and rich animations such as scroll-triggered effects.
-
Visual-to-Code Generation and Debugging
- Kimi K2.5 has made groundbreaking advancements in visual-to-code generation and debugging.
- For example, the model can reconstruct a website from a video and use code to mark the shortest path, showcasing its powerful capabilities in visual reasoning and code generation.
-
Multimodal Tool Invocation
- Kimi K2.5 supports up to 1,500 tool invocations, enabling it to handle high-density, large-scale task flows.
-
Kimi Code Integration
- Kimi K2.5, when integrated with Kimi Code, can work in the terminal and supports integration with various IDEs such as VSCode, Cursor, and Zed.
- Kimi Code supports images and videos as inputs and can automatically discover and migrate existing skills into the working environment.
Application Scenarios and Performance
- Agentic Coding: Kimi K2.5 performs exceptionally well in multiple agentic coding benchmarks, particularly in front-end development.
- Visual Reasoning: The model can handle complex visual reasoning tasks, such as reconstructing a website from a video and marking the shortest path.
- Agent Swarm Collaboration: In internal evaluations, Kimi K2.5's agent swarm technology reduced the end-to-end runtime of complex tasks by 80% and supported more complex, longer-horizon workloads.
Industry Impact and Developer Recommendations
The release of Kimi K2.5 marks a significant advancement in the field of agentic intelligence, providing developers with a more powerful tool to handle complex AI tasks. Developers can leverage Kimi K2.5's agent swarm technology to build more efficient and intelligent AI applications, especially in scenarios that require multi-step tool invocation and complex task coordination.
Conclusion
The release of Kimi K2.5 not only demonstrates Kimi's continuous innovation in AI but also solidifies its leadership in the field of agentic intelligence. As Kimi K2.5 becomes widely adopted, the capabilities of AI in visual reasoning, code generation, and agentic collaboration will be further enhanced.
— END —Source: Kimi Blog (2026-09-07)
Tags: #Kimi #Open-Source Model #Agentic Intelligence #Visual Reasoning #Multimodal
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments