On July 27, 2026, the Chinese company Moonshot AI is opening the full weights of the Kimi K3 model — the largest open AI model in the world. We break down what this model is, how to use it for free right now, and why within just hours after release it surpassed top models in one of the key rankings.
What Is Kimi K3
Kimi K3 is the flagship model from Moonshot AI, an artificial intelligence lab from Beijing that develops the Kimi AI assistant. The model was released on July 16, 2026, and immediately gained the status of the largest open language model in history — 2.8 trillion parameters built on a Mixture-of-Experts architecture (only 16 of the 896 “experts” inside the model are activated for each request, making computation more efficient).
The model has a context window of 1 million tokens — this means it can process entire code repositories, large legal archives or several full books in a single request without losing the thread between parts of the text.
How to Try Kimi K3 for Free
Since July 19, access to Kimi K3 has been completely free and does not require registering a developer account (although there are limits on the number of requests). You can try the model in several ways:
- Through the Kimi app on iOS and Android
- On the kimi.com website
- Through the Kimi Work desktop app
- Through the developer API at platform.kimi.ai
Why Kimi K3 Surpassed Claude Fable 5 in One Ranking
The loudest moment of the launch: literally within a few hours after release, Kimi K3 rose from 18th to 1st place in the Frontend Code Arena ranking, scoring 1679 points and overtaking Anthropic’s Claude Fable 5 — the model that most developers had previously considered unreachable specifically in frontend development.
It is important to add context: in overall performance across a broad set of tasks, Kimi K3, according to Moonshot’s own data, still trails both Claude Fable 5 and Sol from GPT-5.6. In other words, this is not about complete leadership, but about a win in a specific, narrow category that is important for developers.
Full Open Weights — the Main Event of July 27
Today, July 27, Moonshot AI is releasing the model’s full open weights. In practice, this means that anyone with sufficiently powerful hardware will be able to download the complete model, run it on their own servers and even fine-tune it for their own tasks — without depending on Moonshot’s cloud infrastructure or any other company.
This is a fundamental difference from closed models such as GPT-5.6 or Claude Fable 5, which can only be accessed through the official APIs and interfaces of the developer companies.
In Brief: The Key Points
- July 16, 2026 — Kimi K3 release, 2.8 trillion parameters, the largest open model in the world
- July 19 — free access for everyone, without developer registration
- July 27 — release of the model’s full open weights for self-hosting
- Context window — 1 million tokens
- Within hours, it surpassed Claude Fable 5 in the Frontend Code Arena ranking (1679 points, 1st place)
- In overall performance, it still trails Claude Fable 5 and GPT-5.6 Sol
FAQ
What is Kimi K3? The flagship AI model from the Chinese company Moonshot AI, the largest open language model in the world — 2.8 trillion parameters.
Can Kimi K3 be used for free? Yes, since July 19, 2026, access has been free through the app, the kimi.com website or the desktop version, without registering a developer account, with limits on the number of requests.
How is Kimi K3 better than Claude Fable 5? In the narrow category of frontend development, the model surpassed Claude Fable 5 in the Frontend Code Arena ranking. In overall performance across a broad set of tasks, it still trails Fable 5.
What do “open weights” of a model mean? It means the model parameters are fully available for download — anyone with sufficiently powerful hardware can run the model on their own servers without depending on the developer company’s cloud.
What context window does Kimi K3 have? 1 million tokens — the model can process entire code repositories or several books in a single request.
The article was prepared by the TechVisor team — practical IT media for people.
