Moonshot AI is a Chinese AI startup focused on developing advanced large language models, most recently releasing its Kimi K2 model, which is designed for complex "agentic" tasks and has enhanced coding and reasoning abilities. The company is also known for its cost-effective development strategy, training models using older chips like the Nvidia H800, and its open-source approach, making its models available to the broader developer community.
Key activities and achievements
Developing the Kimi K2 model: The company released Kimi K2 Thinking, a new open-source model with enhanced reasoning and agentic capabilities that can autonomously use tools to solve complex problems. It has a 1-trillion-parameter architecture.
Focusing on agentic capabilities: Kimi K2 is designed to outperform other models in "agentic" tasks, which are the ability to understand and act on complex, open-ended user requests with less explicit instruction.
Demonstrating cost-efficient development: Moonshot AI trains its advanced models using older hardware like Nvidia's H800 chips, which were previously exported to China, to show resilience despite U.S. export controls on newer chips. The training for Kimi K2 cost an estimated $4.6 million.
Advocating for open-source models: By releasing Kimi K2 as an open-source model, Moonshot AI aims to broaden developer involvement, foster global innovation, and compete in the AI market.
Offering competitive pricing: The company offers its models with a lower cost structure for developers compared to competitors like OpenAI and Anthropic.
Fundraising and expansion: Moonshot AI is in the process of raising significant funds, potentially valuing the company at around $4 billion, and plans to pursue an IPO to further expand its operations.
“China's Moonshot AI is nearing a dollar funding round that could boost its valuation to around $4 billion, according to people familiar with the matter.
The Beijing-based AI startup is in talks with global investors, including investment firm IDG Capital, for the fundraise, which could total several hundred million dollars, some of the people said
Other potential investors include existing shareholder Tencent -- one of China's biggest tech conglomerates -- some of the people said.
Moonshot aims to close the funding round by the end of this year, some of the people said. It has also told some potential investors that it will aim for an initial public offering in the second half of next year, according to one of the people.
A representative from Moonshot said the company hasn't given any IPO timeline to investors.
"We have not communicated any specific IPO timeline to investors, and the details in the report do not reflect our plans or statements," the representative said.
IDG didn't respond to a request to comment. Tencent declined to comment.
Moonshot, which has gained recognition for its long-form text analysis, has already attracted a number of prominent backers, including Chinese tech giant Alibaba Group and venture-capital firm HongShan Capital.
In August last year, it raised more than $300 million in a Series B round from investors including Tencent and Gaorong Capital, another VC firm, according to people familiar with the matter.
Moonshot is one of the rising stars in China's thriving AI sector, where more firms are making breakthroughs in technological advancements and moving toward public listings.
Shanghai-based MiniMax, which also counts Tencent as an investor, has confidentially filed for a Hong Kong listing, The Wall Street Journal reported in July.
Beijing-based Zhipu AI has also been preparing for an IPO.
Chinese AI startups are valued at a fraction of their U.S. counterparts. OpenAI, for example, was recently priced at $500 billion in its latest round, the Journal reported last month.
Still, since DeepSeek first burst onto the scene in January this year with a low-cost model that could rival those by Western counterparts, the momentum in China's homegrown AI scene has been building. China's established players are joining in too, with big names like Alibaba also vying for an edge in the AI race.” [1]
How long are texts that Moonshot AI takes to work with for free?
Moonshot AI does not have a "free plan" with a text length limit, but its models support varying context windows depending on the specific model version, ranging from 8K to 128K tokens.
Free usage is available through third-party platforms like Kimi-K2-Thinking on Medium. For the best performance, these models require significant computational resources, which are not necessary for the free version.
Free usage
You can use a version of Moonshot AI's model for free through platforms that offer it, such as Kimi-K2-Thinking on Medium.
This is distinct from the paid API, which has different models with varying context windows.
Model context windows
The moonshot-v1-8k model has a maximum context of 8,192 tokens.
The moonshot-v1-32k model has a maximum context of 32,768 tokens.
The moonshot-v1-128k model has a maximum context of 128,000 tokens.
Other considerations
The context window is the total number of tokens for both the prompt and the response.
The latest Kimi K2 model, while not having a "free" version, supports up to 256,000-token contexts.
1. China's Moonshot AI Set to Raise Funds at Near $4 Billion Value. Qu, Tracy; Huang, Raffaele. Wall Street Journal, Eastern edition; New York, N.Y.. 25 Nov 2025: B4.
Komentarų nėra:
Rašyti komentarą