01What is Kimi K2.7 and how does it work?

Kimi K2.7 is an open-source AI model from a small Chinese lab called Moonshot AI. It can build complete apps, games, and websites from a single sentence in plain English. You don’t need to be a coder to use it.

The model uses a 1 trillion parameter mixture of experts architecture. Only 32 billion parameters are active per token, which keeps it fast and cheap. It has a 256k context window, so it can handle very long tasks.

It is built for long agentic coding tasks, meaning it can code on its own for hours without stopping.

You can run Kimi K2.7 for free using Ollama. It is open source, so anyone can use it. In this episode, Julian shows how it can build an entire agent operating system, create games, and even produce fully edited videos with an AI avatar.

02Main features of Kimi K2.7

Kimi K2.7 has several standout features that set it apart from other AI models.

  • Open source: Anyone can use it for free, and it runs on Ollama.
  • Long agentic coding: It can work on tasks for hours by itself, making it great for complex projects.
  • Coding plan: It has a built-in coding plan, so you can plug it directly into tools like Minix and Hermes.
  • Parallel task handling: It can handle multiple tasks at once, such as coding an app while editing a video.
  • Plain English input: You describe what you want, and it writes the code for you.
  • Self-checking: It can check its own work and iterate until it meets quality standards.

These features make it powerful for building tools, websites, and even full games with minimal effort.

03How to get started with Kimi K2.7

  1. Open Ollama on your computer and make sure it is installed.
  2. Search for Kimi K2.7 in Ollama’s model library and pull it.
  3. Open Kimi Code (or the interface you prefer, like Hermes) and start a new chat.
  4. Write your prompt in plain English. For example, “build a stock tracker for my small shop.”
  5. Open the workspace to see the app or tool running.
  6. If you want long autonomous work, set up a coding plan or connect it to an agent operating system.

That’s it. You don’t need to write any code yourself. Kimi K2.7 does the coding, checking, and running for you.

04How to use Kimi K2.7 for real projects

Julian shows many examples of what you can build with Kimi K2.7. In one, he created a full agent operating system—a dashboard with different apps. When you type something into it, it saves automatically.

You can close it, open it again, and it works like Windows, but simpler.

He also built a full game inside the agent operating system, including an open-world game. Everything you see in the game was coded by Kimi K2.7 itself. It did this in a couple of hours, working in parallel on different parts.

Another example is a fully edited video with an AI avatar. Julian gave Kimi K2.7 a prompt and the skill to create videos. The model wrote the script, edited the video, and created the avatar—all without Julian’s help.

To get quality results, Julian recommends using a Kanban board with a team of agents. For example, you can have a video judge agent that scores the video. If the score is below 8 out of 10, it sends the video back for improvements.

This loop runs until the video meets the standard.

05Benchmarks and comparisons: Kimi K2.7 vs Claude Opus 4.8

Kimi K2.7 performed well on benchmarks, but most of them came from Moonshot AI itself. Treat them as direction, not gospel.

In one independent test, Kimi K2.7 beat Claude Opus 4.8 on a tool use benchmark. The test measures whether the AI can use other tools to get a job done—like opening a file, making a change, or updating a ticket. Kimi scored 81.1, while Claude scored 76.4.

However, another researcher tested Kimi K2.7 against an older version and another top model. His take: the new version is more honest in how it writes code, but not actually more capable on his tests. On one task, the new version scored worse than the old one.

Julian compared Kimi K2.7 to GLM 5.2, another Chinese model. He prefers GLM 5.2 for general coding because it feels more like Claude. But for agentic tasks, Kimi K2.7 replies faster when used with Hermes.

06Tips, limits, and mistakes to avoid with Kimi K2.7

Here are some practical tips from Julian to get the best out of Kimi K2.7.

  • Use a system to store your work: Without a proper place to save your projects, you will lose them. Julian lost a year of work this way. Use an agent operating system or similar tool.
  • Accept its limits: On the hardest, most complex jobs, expensive closed models still lead. Kimi is best for simple tools, forms, trackers, and small apps.
  • Don’t trust benchmarks blindly: Since Moonshot AI graded its own homework, take those results with a grain of salt.
  • Don’t fear the name “code”: You don’t need to be technical. Just describe what you want in plain English.
  • Use a quality checker: Set up a judge agent to review outputs and send them back if they don’t meet your standards.

These tips help you avoid common mistakes and get useful results.

07Who should use Kimi K2.7 and why it matters

Kimi K2.7 is for anyone who has ideas but lacks coding skills. It is especially useful for small business owners who need simple tools, websites, or customer features. For example, a small shop owner could describe a stock tracker and have it built before launch.

An online seller could create a tool that tells customers delivery dates.

The real power of Kimi K2.7 is not just the model itself—it’s how you combine it with other tools. Julian has built an agent operating system that puts Kimi, Claude, Hermes, and other models in one dashboard. They share one memory and work together.

You can even build a full company of AI agents with Kimi and Hermes.

This matters because AI is moving fast. The same lab that made Kimi K2.7 launched its first big model in July last year. Since then, it has shipped five major versions in under a year.

Tasks that used to take years are now done in months or weeks. People who start describing their ideas into reality today will be ahead.

08How Kimi K2.7 fits into the bigger picture

Kimi K2.7 shows how AI is becoming cheaper and more accessible. It costs almost nothing to run, and it’s open source. This is a big change from just a few years ago when powerful AI was expensive and closed.

Julian notes that the Fable situation—where a similar tool got banned—has made people look for alternatives. Kimi K2.7 dropped at the perfect time to fill that gap. It can build things like an operating system, games, and videos, all from one sentence.

The speed of improvement is the key thing to watch. Moonshot AI went from zero to five major versions in a year. That means what Kimi K2.7 can do today will seem basic soon.

The people who adapt to this speed will win.

If you want to try it, Kimi K2.7 is available now on Ollama. Start with a simple project and see what it can do.