Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Chinese companies continue to produce AI models that rival the capabilities of the systems developed by OpenAI and other US AI companies.
This week, MiniMaxThe startup of Alibaba- and Tencent that is he was promoted approximately $850 million in capital and is valued at more than $2.5 billion, first appearance three new models: MiniMax-Text-01, MiniMax-VL-01, and T2A-01-HD. MiniMax-Text-01 is a text-only format, while MiniMax-VL-01 can understand images and text. The T2A-01-HD, meanwhile, produces audio — mostly audio.
MiniMax claims that MiniMax-Text-01, which is 456 billion units in size, outperforms models such as Google recently revealed. Gemini 2.0 Flash on benchmarks like MATH and SimpleQA, which test a model’s ability to answer math and truth-based questions. Parameters correspond to the model’s ability to solve problems, and models with more parameters perform better than those with fewer parameters.
As for the MiniMax-VL-01, MiniMax says it competes with Anthropic Claude 3.5 Sonnet for analysis that requires multi-dimensional understanding, such as ChartQA, which works to answer graph- and image-related questions (for example, “What is the maximum value of the orange line on this graph?”). In fact, the MiniMax-VL-01 does not outperform the Gemini 2.0 Flash in most of these tests. It’s OpenAI GPT-4o and Meta Rule 3.1 hit a few more.
Note, MiniMax-Text-01 has a very large window. A model’s context, or information window, defines the inputs (such as words) that the model considers before outputting them (additional words). With a screen that holds 4 million tokens, the MiniMax-Text-01 can scan nearly 3 million words at a time – or more than five copies of “War and Peace.”
For context (no pun intended), the MiniMax-Text-01 window is about 31 times the size of GPT-4o’s and Llama 3.1’s.
The latest MiniMax model released this week, the T2A-01-HD, is an audio recorder that is designed for speech. The T2A-01-HD is capable of producing voice-overs with clear voice, tone, and sound quality in 17 different languages, including English and Chinese, and output audio from just 10 seconds of recording.
MiniMax did not publish results comparing the T2A-01-HD with other audio models. But to this reporter’s ear, the T2A-01-HD’s output sounds similar to audio from Measure and basics like PlayAI.
Apart from the T2A-01-HD, which is only available through MiniMax’s API and Hailuo AI platform, new MiniMax models can be downloaded from GitHub and the AI dev platform Hugging Face.
Just because the models are “transparent” there doesn’t mean they aren’t locked into other things, however. MiniMax-Text-01 and MiniMax-VL-01 it’s not really open in the sense that MiniMax did not generate the components (ie, the training data) needed to be rebuilt from scratch. In addition, they are under MiniMax’s restrictive license, which prevents developers from using these models to improve AI models and requires platforms with more than 100 million monthly users to request a special license from MiniMax.
MiniMax was founded in 2021 by former employees of SenseTime, one of China’s largest AI companies. The company’s services include apps like Talkie, an AI-powered platform Character AIand examples of texts and videos that MiniMax has produced in Hailuo.
Some of the MiniMax products have been the subject of a little controversy.
Talkie, which was pulled from Apple’s App Store in December for undisclosed “technical” reasons, features AI avatars of people, including Donald Trump, Taylor Swift, Elon Musk, and LeBron James, none of whom appear to have agreed to appear. program.
In December, Broadcast magazine report that MiniMax video producers can reproduce the logos of British British cinema, showing that the MiniMax models were trained on the content of the channel. And MiniMax says they are being prosecuted by iQiyi, a Chinese video streaming service that claims MiniMax was illegally trained on iQiyi footage.
The new MiniMax models will arrive after the days of Biden’s administration have come to an end provided strict export regulations and restrictions on AI technology for Chinese businesses. Companies in China have already been banned from buying AI chips, but if the new rules go into effect as written, companies will face tougher tools in semiconductor technology and the types needed to launch advanced AI systems.
On Wednesday, the Biden administration he announced some that focus on stocking high-end chips from China. Chip companies that want to export some chips will get a bigger license if they have to check and make sure their products don’t reach Chinese customers.