Kimi K3.1 Pro: the Chinese AI that beats Claude in some tests
Kimi K3.1 Pro: the Chinese AI that beats Claude in some tests
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is the new AI model from China's Moonshot AI, and in some tests it beats both Claude and OpenAI's most expensive model. It is also cheaper to use. But when independent testers worked out what it costs in real work, the picture became more mixed.
Kimi is Moonshot's answer to ChatGPT: a chat service on the web and on your phone. Alongside it come Kimi Work for office tasks and spreadsheets, and Kimi Code for programming. The previous version, Kimi K3, was already one of the best open source AI models. K3.1 Pro is a clear step up.
Where Kimi K3.1 Pro leads — according to Moonshot
Moonshot highlights three areas where Kimi beats the competition:
- Hard knowledge questions. On one of the toughest knowledge exams around, with expert questions from hundreds of subjects, Kimi scores 70 percent. Claude Opus 5.5 scores 67.7 and GPT-6 Astra 57.2.
- Long programming jobs that run for hours: Kimi gets 51.7 points against 46.3 for Opus and 42.5 for Astra.
- Finance and office work, such as reading reports and answering questions about the numbers: 64.4 against 62.2 for Opus.
For a business, something else may matter more: Kimi makes things up less often. According to Moonshot, the share of invented answers dropped from 23.4 to 8.3 percent compared with the previous version. In MindStudio's tests it also said honestly when it hadn't finished, instead of pretending.
Independent tests are more cautious
The figures above come from Moonshot itself. According to Artificial Analysis, which compares AI models independently, Kimi K3.1 Pro ranks fourth in the world with a score of 55. Claude Opus 5.5 scores 58.
In an independent programming test Kimi came sixth out of eleven, behind Opus, Astra and Anthropic's budget model Haiku 5.5. Moonshot's claim that Kimi beats Astra by seven points did not hold up there.
The fourth-best AI in the world is still very good. But it is not the clear leader the press release suggests.
Kimi Code, Moonshot's app for programming. Screenshot: Rodrigo Padula, CC BY 4.0, via Wikimedia Commons.
What does Kimi K3.1 Pro cost?
On paper, Kimi is cheap. If you build your own solution that talks directly to Moonshot, you pay $2 to send in roughly 750,000 words and $10 to get the same amount of text back. That is half the price of Claude Opus 5.5 and a fifth of GPT-6 Astra. A faster version costs twice as much.
But MindStudio had Kimi solve six programming tasks, and the bill came to $32. Haiku 5.5 solved the same tasks better for $7.40. DeepSeek V4.1 Flash did almost as well for under $2.
The reason is that Kimi rereads the same material over and over. Each read is cheap, but it reads a lot. It is like a tradesperson with a low hourly rate who needs twice as many hours.
Should your business use Kimi?
Kimi K3.1 Pro is an interesting option, but not for everyone:
- Try it if you do a lot of knowledge-heavy analysis, such as finance, reports or research. That is where it is strongest.
- Compare first if you pay per use for big jobs. Cheaper models are often enough.
- Choose a flat monthly price if you use it a lot. According to MindStudio, Kimi's own subscription works out cheapest for heavy use.
- Think about what you send. Moonshot is a Chinese company and whatever you type ends up on its servers.
Moonshot has also promised to release the model freely, so anyone can run it. But it is so large that it takes a full server rack, so for most small businesses that is not a realistic option.
What can we learn from this?
Kimi K3.1 Pro shows that the top of the AI world is no longer only American. But it also shows why you should read a maker's own figures with some scepticism. What counts is how the model handles your tasks, and what the bill looks like afterwards.
Want to know which AI model suits your business best? Talk to us and we will help you test and compare before you decide.
Frequently asked questions
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is an AI model from the Chinese company Moonshot AI and the successor to Kimi K3. It powers the Kimi chat service, Kimi Work for office tasks and Kimi Code for programming, and can also be connected to your own solutions. It reads both text and images and can hold very large documents in memory at once.
How much does Kimi K3 cost?
Kimi K3.1 Pro costs $2 per million input tokens and $10 per million output tokens, where a million tokens is roughly 750,000 words. That is half the price of Claude Opus 5.5 and a fifth of GPT-6 Astra. The faster K3.1 Pro High Speed costs $4 and $20. In practice the bill can still be high, because the model rereads a lot of material.
Is Kimi better than Claude?
In some tests. According to Moonshot, Kimi K3.1 Pro beats Claude Opus 5.5 on hard knowledge questions (70 versus 67.7 percent), long programming jobs and financial office tasks. In independent comparisons Claude Opus 5.5 is still ahead: Artificial Analysis gives Opus a score of 58 against Kimi's 55, and in an independent programming test Kimi came sixth out of eleven.
Is Kimi AI safe for businesses to use?
Kimi is run by Moonshot AI, a Chinese company, and what you type is processed on its servers. Businesses should therefore avoid sending personal data, customer data or trade secrets there without first reviewing the terms. Moonshot has promised to release the model freely so it can run on your own servers, but that requires very powerful hardware.
Who is behind Kimi?
Kimi is developed by Moonshot AI, an AI company based in Beijing. Its models are among the strongest from China, alongside DeepSeek, Alibaba's Qwen and Z.ai's GLM. The predecessor Kimi K3 is available under an open licence, and Moonshot plans to release K3.1 Pro the same way.