5Y News｜Moonshot AI Releases and Open-Sources K2.5 Model, Bringing New Visual Understanding, Coding, and Agent Clustering Capabilities

On January 27, 2026, Moonshot AI released and open-sourced the Kimi K2.5 model:

• It is Kimi's most intelligent model to date, achieving open-source state-of-the-art performance on Agent, coding, image, video, and a range of general intelligence tasks.
• It is also Kimi's most capable model yet, built on a native multimodal architecture that simultaneously supports visual and text inputs, thinking and non-thinking modes, and both conversational and Agent tasks.

We believe that smarter, more capable models with stronger coding abilities will help advance technological equity and benefit more people:

• Kimi K2.5 makes intelligence more accessible. By combining visual understanding with reasoning, coding, and Agent capabilities, K2.5 lowers the barrier for users to interact with AI: when words fall short, you can simply take a photo, screenshot, or screen recording and send it to Kimi, breaking through the limitations of text-based expression.
• Kimi K2.5 helps everyone master Office. The K2.5 model extends Kimi's Agent capabilities into daily office work, beginning to master intermediate and advanced skills in Word, Excel, PowerPoint, PDF, and other common software — helping users directly deliver near-professional quality office documents.

We believe AI Agents will give everyone "superpowers." Just as human society requires collaboration to tackle truly complex work, K2.5 introduces "Agent Clusters" for the first time, enabling K2.5 to autonomously create its own "avatars," assembling teams of different roles on demand to work in parallel — achieving 10x or even 100x efficiency gains.

Kimi K2.5 is now available on kimi.com, the latest Kimi App, the Kimi API open platform, and our programming assistant product Kimi Code. We welcome you to try it out.

For more technical details and benchmark results on the Kimi K2.5 model, please follow Kimi's technical blog and upcoming technical reports.

Code × Vision

Making Intelligence Accessible

Kimi K2.5 further elevates the coding capabilities of open-source models, especially in frontend development. The K2.5 model supports generating complete frontend interfaces from simple natural language conversation, and effectively handles interactive layouts, scroll-triggered animations, and other dynamic effects. Below are examples of website development achieved by K2.5 with the assistance of an image generation tool, using only a single prompt:

By integrating visual capabilities, K2.5 truly lowers the barrier to programming: you can simply upload a screen recording. Kimi K2.5 automatically breaks down the underlying interaction logic and reproduces it from start to finish with clean, professional code.

The advanced visual understanding and programming capabilities of the Kimi K2.5 model have also been validated by feedback from early API beta testers:

Among them, Keep AI's coach Kaka, built on Kimi K2.5's video action recognition and assessment capabilities, will be launching soon — stay tuned.

Agent Clusters

Giving You "Superpowers"

Six months ago, Kimi released the first trillion-parameter open-source Agent model, Kimi K2. Subsequently, Kimi K2 Thinking was introduced, which gained the ability to independently complete long-horizon tasks of up to 300 steps by increasing thinking time. But that wasn't enough. Solving real-world complex problems requires more than going solo — it demands teamwork.

Today, we unveil a new exploration for Kimi K2.5: Agent Clusters. This time, we've evolved from a single Agent to Agent Clusters.

When facing complex tasks, K2.5 is no longer an all-encompassing "jack-of-all-trades expert," but transforms into an instantly assembled "professional team." It can dispatch up to 100 avatars on the fly to process 1,500 steps in parallel, all based on task requirements. All role assignments and task decomposition are decided on the spot by K2.5 itself, with no pre-configuration needed.

Let's look at an example. Feed the Kimi Agent Cluster 40 papers on psychology and AI. Kimi first reads through all 40 papers sequentially via multiple tool calls, ensuring that all necessary information is fully preserved in the context. It then spawns several sub-agents — essentially Kimi's "avatars" — each responsible for writing different sections. Finally, the main agent acts as quality control, consolidating everything into a professional PDF review dozens of pages long:

While K2.5 has already reached advanced levels on mainstream Agent benchmarks, we care more about every minute it saves users. In large-scale search (wide search) scenarios, compared to single-Agent execution, Agent Clusters reduce the minimum critical steps needed to achieve target performance by 3 to 4.5x, with savings increasing further as target requirements rise; through parallelization, actual wall-clock time can be shortened by up to 4.5x:

Moreover, scaling training for Agent Clusters is extremely challenging. To address this, we rebuilt our reinforcement learning infrastructure and specifically optimized training algorithms to ensure peak efficiency and performance. This experimental feature is now in Beta testing and will be gradually rolled out soon. We look forward to Kimi K2.5's collaborative capabilities helping you tackle even more difficult problems.

Kimi Code Officially Launched

Since the release of the Kimi K2 series models, they have been well-received by developers at home and abroad for their outstanding performance in software engineering. Backend data from the Kimi open platform shows that a large number of developers are pairing Kimi K2 series models with coding agent products like Claude Code, Cline, Roo Code, and Kilo Code. At the same time, coding agent products are increasingly demonstrating more general-purpose capabilities, with their user base expanding beyond technical professionals — suggesting enormous future potential.

Today, we officially introduce Kimi's programming tool: Kimi Code. It runs directly in the terminal and integrates seamlessly with mainstream editors including VSCode, Cursor, JetBrains, and Zed. Kimi Code fully leverages K2.5's multimodal advantages, supporting direct image and video input for programming assistance, and can automatically discover and migrate your existing skills to new workflows.

Kimi Code Bench is our internal code capability evaluation benchmark, covering various end-to-end tasks from building, debugging, refactoring, and testing to scripting, supporting multiple programming languages. In our evaluation, Kimi Code powered by K2.5 showed substantial improvements over previous Kimi models.

We welcome you to use the Kimi K2.5 model API together with Kimi Code, or access it through Kimi's monthly membership plan (kimi.com/code). Additionally, the Agent SDK behind Kimi Code will also be released as open source, helping everyone customize their own Agent experience. We've provided more information on GitHub — details at https://github.com/MoonshotAI/kimi-agent-sdk/tree/main/examples.

5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing them with support ranging from the spiritual to all operational aspects of running a business. We believe that if the world begins to believe in the "crazy" you that others see, the world will become a different place.

BEIJING · SHANGHAI · SHENZHEN · HONG KONG

WWW.5YCAP.COM