Kimi by Moonshot AI: The Chinese AI Provider Overview

Anyone getting an overview of the major artificial intelligence providers has, since 2025, increasingly come across a name that was barely known in Germany until recently: Kimi. Behind the chatbot and its associated model series is the Chinese company Moonshot AI. For case workers, real estate agents or property managers wondering whether it’s worth looking beyond the established providers like OpenAI, Anthropic or Google, Kimi is an interesting but also a particularly complex case. That’s because Moonshot AI pursues an unusual dual strategy: on one hand, the company releases its base models openly and for free download; on the other, the convenient chat and API access runs via servers in China. Anyone processing sensitive business data, such as tenant data or contract documents, should know this difference before using Kimi in their everyday work – placed in context within our overview of AI models compared.

Who is behind Kimi? The history of Moonshot AI

• Kimi website
• Moonshot AI website

Moonshot AI was founded in Beijing in March 2023. Among its founders is Yang Zhilin, an AI researcher with a doctorate from Carnegie Mellon University who had previously co-authored well-known language model papers such as XLNet and Transformer-XL. Together with other co-founders, he built a company that Chinese media often mention in the same breath as ByteDance and OpenAI – as an example of a particularly ambitious AI startup.

Financially, Moonshot AI has been backed by, among others, the technology corporations Alibaba and Tencent as well as investors such as IDG Capital. Its company valuation rose considerably within a short time: from around 300 million US dollars at founding, to about 2.5 billion US dollars in early 2024, up to figures in the multi-billion US dollar range reported for early 2026. This makes Moonshot AI one of the most highly valued private AI companies in China.

From niche chatbot to internationally recognized model

Kimi initially became known as a Chinese-language chatbot. Its international breakthrough came in summer 2025 with the Kimi K2 model series, introduced as an open model with a very large parameter count, which caused a stir in expert circles because it could compete with considerably more expensive models from Western providers on benchmarks for programming tasks and agentic capabilities.

Feature	Detail
Company	Moonshot AI (Beijing Moonshot Technology Co., Ltd.)
Founded	March 2023, Beijing
Founder	Yang Zhilin and co-founders
Key investors	Alibaba, Tencent, IDG Capital, other venture capital firms
Best-known model	Kimi K2 (since July 2025), further developed up to K3 (2026)
End-user product	Kimi chatbot and Kimi app (kimi.com)
License model for base models	Open weights under a modified MIT license

Lukinski AI · ab 49 €/MonatFrag die Immobilien-KI — Jetzt testen →Frag z.B.: „Ist diese Mietvertragsklausel wirksam?“

Kimi as a product: chat app, agent mode and API

For most users, Kimi is first and foremost the web and app interface at kimi.com. There, the model can be used like an ordinary chatbot for texts, translations, research or programming tasks. Since autumn 2025, Moonshot AI has additionally offered an extended agent mode, described in reports under names such as “OK Computer” or “Kimi Work.”

Agent mode

In this mode, Kimi can independently take on multi-step tasks: it can research on the web, create spreadsheets and simple multi-page websites, run Python code, and, according to Moonshot AI, read in several dozen files at once. For a case worker in property management, this could theoretically mean that a larger amount of documents – such as several utility bill statements or listing exposé drafts – is summarized in one pass. Important here: all uploaded content is processed on Moonshot AI’s servers, more on that below.

Kimi via the API

Developers and companies can also access the Kimi models via their own programming interface (API), for example to integrate them into existing software such as a CRM system or document management. This interface is offered directly by Moonshot AI as well as by several third-party providers, through which the open model weights can also be run on your own or European cloud infrastructure. This is the decisive difference from pure chat use: anyone who self-hosts the open weights also decides for themselves where the data is stored.

Rule of thumb: Kimi is not “one” product, but two – the convenient chat and agent access with data processing in China, and the openly available model weights, which can in principle also run on your own infrastructure.

Pricing models: what does Kimi cost?

For private use via the chat interface, Moonshot AI offers free basic use with restrictions on request volume and feature scope, supplemented by paid packages for more capacity. For developers and businesses, API billing by text volume (tokens) is the most relevant. The following figures reflect the publicly available price lists as of August 2026 – as with all AI providers, these prices can change on short notice, so it’s always worth checking Moonshot AI’s current pricing page before doing a business calculation.

Model version	Approximate price per 1 million input tokens	Approximate price per 1 million output tokens
Kimi K2.5 / K2.6 (as of 2026)	approx. 0.60-0.95 US dollars	approx. 3.00-4.00 US dollars
Kimi K3 (from July 2026)	approx. 3.00 US dollars	approx. 15.00 US dollars

Practical example: a property management company has 500 utility bill statements automatically summarized.

• 1,500 words per statement
• Around 750,000 words in total
• About 1 to 1.3 million input tokens
• K2 series: under 2 US dollars for input processing
• K3: correspondingly higher costs

On top of the input comes the usually more expensive output – the summary created.

Data protection and GDPR: what does the server location China mean?

For use at a German business, this point is the most important one: if you use Kimi via the website kimi.com or via Moonshot AI’s official API, your inputs are processed on servers in China. China currently has no so-called adequacy decision from the European Commission – this means the EU does not officially classify the level of data protection there as equivalent to the General Data Protection Regulation (GDPR). Similar concerns are also discussed with other Chinese AI providers such as DeepSeek.

According to Moonshot AI’s publicly available terms of use, entered content can also be used to further develop its own models. For purely private, non-binding use, that may be an acceptable risk. For a case worker who wants to process personal data from a lease, a credit report, or a customer inquiry, however, the starting position is different: here, the strict GDPR requirements for data processing apply, and transferring such data to a provider headquartered and running servers in China can hardly be made legally sound without additional contractual and technical safeguards.

Rule of thumb: anyone using Kimi for general research or non-binding text drafts without personal data involved takes on a different risk than someone entering tenant data, names, or contract content. The latter should be avoided without additional safeguards.

A conceivable alternative for privacy-sensitive use cases is running the open Kimi model weights on your own or European server infrastructure, so that the data never has to leave the country – however, this requires your own technical expertise and corresponding hardware or cloud costs. For businesses in the real estate industry looking for an AI solution with German hosting and without this configuration effort, specialized offerings such as Lukinski AI are available as an alternative, designed from the outset for German server locations and the industry’s requirements.

Strengths and weaknesses of Kimi

Strengths	Weaknesses
Open model weights, making self-hosting possible too	Standard chat and API use with data processing in China
Very large context window in newer versions	No known official offering with EU data residency
Strong results on programming and agentic tasks	German language quality and localization less well documented than with established providers
Comparatively inexpensive API prices for the smaller model versions	Corporate environment (investors, regulation) harder for German businesses to assess

Who is Kimi suited for?

Kimi is particularly well suited to certain groups of users.

• Technically skilled users and developers
• Companies with their own hosting
• Programming and automation tasks

For real estate agents with listing exposé texts or property managers with correspondence, caution is advised as soon as personal data is involved. Here, a comparison with European providers is worthwhile.

Frequently asked questions about Kimi and Moonshot AI (FAQ)

Can Kimi be used for free?

Yes, there is free basic use with restrictions via the chat interface. For higher usage volumes or API integration, costs apply based on the amount of text processed.

Is Kimi safe for business or personal data?

When using Moonshot AI’s standard chat or API access, data is processed in China, for which there is no EU adequacy decision. Special caution is therefore advised for personal or contractually sensitive data from the real estate or finance sector.

What distinguishes Kimi from ChatGPT, Claude or DeepSeek?

Kimi differs from most Western providers mainly through the open release of its model weights. There are clear parallels with DeepSeek, another Chinese provider, regarding data protection issues and server location.

Can Kimi also be run without Chinese servers?

In principle yes, since the base models are released as open weights and can also be made available on European infrastructure via various third-party providers. However, this requires technical expertise and your own server capacity.

What does the model name K2 stand for?

K2 refers to the large model generation from Moonshot AI introduced in summer 2025, named after the second-highest mountain on Earth. Details on the individual K2 versions and their successors can be found on the in-depth models page.

More technical details on the individual model versions can be found on our page Kimi models in detail. A comparison of all ten AI providers covered in the cluster is available on the overview page.