In actual testing, the Kimi K2.5 is the closest domestic open-source model to the Gemini 3 Pro.

Large ModelIn this era, code is indeed becoming increasingly cheap. Big data pushes out a ton of data every day. AI After looking at the generated web pages for a while, I really get a bit tired of them.

These web pages always have a strong... AI The same old rounded-corner cards, card shadows, Tailwind purple, emojis…

It was posted by Kimi just yesterday. K2.5 The models are all over social media; this time, that's what they're focusing on.Visual programmingA screenshot or a video is all it takes to replicate an almost identical webpage.

Today, let's put the Kimi K2.5 to the test and see if its "vision" and "taste" are up to par.

First, we sent Kimi a screenshot of the nice template.

Prompt wordsverySimpleTo replicate this website.

It's truly amazing. You can see that K2.5 can recognize and extract visual elements from the screen and convert the images into interactive web page code.

MoreoverautomaticA responsive layout has been implemented, allowing you to directly switch between mobile and desktop web browser previews.

And their Agent The generated webpage also has a very user-friendly design. By simply clicking the edit button in the upper right corner, you can directly select the parts that need to be modified in the preview interface and tell Kimi in plain language what to change so that she can make precise adjustments.

K2.5 will automatically deploy the generated webpage for us. Simply click the share button in the upper right corner of the page to copy and access the link.

Replica link: https://rqgcia4tyj2lk.beta-ok.kimi.link

Let's examine the video replication capability next.

Prompt wordsContinue to extendSimpleStyle:Please help me replicate an identical webpage..

The Kimi K2.5 captures the visual features of the video very accurately; the layout, color scheme, and information hierarchy are all reproduced very well, making it look like it was made exactly like the original.AI The flavor is indeed much milder.

The most crucial underlying change in the Kimi K2.5 is its native visual capabilities.You can directly understand the interface structure and details in images and videos..

So next, let's examine visual capabilities in non-programming fields.

This is a photo of Yao'er Hutong in Beijing. Apart from the hutong itself, there are no other prominent landmarks or text prompts in the picture. Let's see what Kimi can tell.

Prompt wordsDetermine where this is.

Kimi recognized that this was an old hutong (alleyway) within Beijing's Second Ring Road and pinpointed a few approximate locations, but it seemed a bit difficult for it to directly name Yao'er Hutong.

Next, let Kimi determine the height of the boy near the car door in the picture.

Kimi answered almost instantly, telling me that the boy's height was approximately between 183 and 188 centimeters, most likely around 185 centimeters.

Gong Jun's official height is 186 cm, which is very close! (I admit, I was a little biased about this answer.)

We increased the intensity, and I found a 25-year-old civil service exam question and sent it to Kimi.

Prompt wordsAnswer this civil service exam question.

After a brief deduction, Kimi quickly grasped the key to the question: the relative positions of identical patterns, thus choosing the correct answer, D.

At this point, the visual capabilities of the K2.5 are quite intuitive.While guessing the location might not be able to pinpoint the exact location, visual reasoning is a strong point, as it can extract useful information from images to support reasoning..

Besides the model itself, I also discovered that Kimi has quietly released a new office assistant app. AgentIt primarily focuses on handling various file types such as PPT, Word, and Excel files needed in daily work.

Let's see if we can end the era of manually typing in Office?

document Agent

Prompt wordsIn the context of the "dual carbon" goals, discuss the practical significance of green and low-carbon development for China's high-quality economic development. Output a 1500-word PDF file, magazine-style layout, two-column format.

It seems there's no need for further description. The writing style matches the magazine's requirements, and the magazine's internal page layout is also beautiful. Currently, this document assistant doesn't seem to have any competitors on the market.

According to the official introduction,Agent Clustering is a completely new exploration in K2.5—from a single... Agent Evolution Agent team.When faced with complex tasks, K2.5 can autonomously create its own "avatars," forming teams with different roles as needed to work in parallel..

So I tested this.Prompt words:

I would like to ask five well-known Chinese artists or IP designers with distinct styles to design five series of ten cover images for WeChat red envelopes themed around the Year of the Horse. I hope the styles will be different.

Kimi K2.5 created 5 designers with different styles and assigned them image generation tasks.

Let's take a look at the generated results:

Among the five styles, the new ink painting style and the traditional ink painting style are quite similar. The overall quality is quite good, the picture is clear, and the pattern is also beautiful. They can be used directly as cover images.

When I try to find past exam papers on my own, it's very difficult to find a complete set at once. Even when I do find a relatively comprehensive collection, I have to pay to download it.

Agent The cluster can also help me package and organize 10 years of past exam questions and their solutions, directly...One-clickDownloading, that's really impressive.

Prompt wordsBatch download of past 10 years' national civil service examination questions and answer explanations.

In addition, data collection and visualization can be done simultaneously.

Prompt wordsThis tool provides a detailed analysis of the GDP and its fluctuations across my country's provinces over the past decade, visualized as a heatmap of China. It defaults to displaying 2025 GDP figures and allows users to switch years.

For example, Consultant Liang was responsible for collecting data, Descartes was responsible for developing an interactive heatmap webpage, and Jingqiu was responsible for integrating the data collected by Consultant Liang into the webpage.

We then obtained a usable heatmap webpage.

Kimi K2.5 focuses on all-around performance Agent The model now possesses visual capabilities, along with enhanced Office and coding abilities, and the capacity for creating unlimited clones. Agent Cluster.

This may mean that,Large ModelIt is moving from "search engine, question answering assistant" to "personal" AI "Assistant" or even "commercial digital employee." The future competition won't just be about parameters and rankings, but about who can handle more tasks for you in real-world work..

Many repetitive mental tasks will be repriced. Content production, basic development, initial design drafts, and data organization—the time costs that were previously built up based on experience—will be flattened out by the model.

Ultimately, it comes down to whether one's aesthetic sense is up to par and whether one's strategy is effective.

Getting things done under your supervision will be the new turning point.

Original link:Kimi releases new K2.5 model, made in China. AI Are we starting to question each other's aesthetics?