Chinese AI startup MiniMax has closed a round of funding valued at a whopping US$250 million, from a consortium of investors including Tencent, in what is the largest single fundraising transaction in the domestic AI space year to date.
The company completed the financing round at reportedly a post-money valuation of US$1.2 billion, making it a unicorn.
Founded in December 2021, MiniMax became one of the fastest-growing AI startups to achieve the unicorn status.
What’s more, it is one of the latest Chinese companies to launch a bid to develop a Chinese equivalent of OpenAI’s ChatGPT.
Unlike other Chinese tech firms that publicly hype their entry into the area of large language model, MiniMax has kept a low profile.
Reuters first broke the story of the startup’s funding round on June 1, prompting a series of media outlets to follow up and speculate on where the firm is headed.
But none made more substantial revelations beyond the sums of the funds raised.
MiniMax already roped in top-tier venture capital funds like IDG, Hillhouse Capital and Yunqi Parnters during previous financing rounds. Reuters reported that Tencent alone even poured US$40 million this time.
cnrobopedia spoke to Yuan Yeyi, one of the co-founders of MiniMax, who confirmed the mega-deal but declined to comment further.
Founded by a group of former employees at SenseTime, one of China’s AI titans, MiniMax has ridden on the wave of a “large language model” frenzy over the past few months.
Its founding team includes Yan Junjie, ex-president of SenseTime and vice president of its research institute.
Media reported that MiniMax has developed three modals of LLM, comprising text-to-visual, text-to-audio and text-to-text, and covering most applications of generative AI.
The company put its LLM to test by running it on a virtual instant messaging tool called Glow it had debuted in November last year.
Users can interact with the model and help it iterate in terms of intelligence and capacity.
Four months after MiniMax launched Glow, its user base approached 5 million.
The LLM innovations by MiniMax also appeared to have won the recognition of many users, as it ranks just behind top-placed Claude from Anthropic in a list released by SuperCLUE, a domestic platform for rating various LLMs based on their capabilities.
According to the list, MiniMax scored 1,188, higher than GPT3.5’s 1,171 — the first time a Chinese-developed LLM has surpassed GPT3.5 in public tests.
Not long after MiniMax rolled out its proprietary LLM, the company applied it in a number of programs run in conjunction with industry partners such as Volcano Engine and Kingsoft WPS, to test and improve the model.
In March, MiniMax opened an API enabling corporate users to access its text-to-audio LLM.