Chinese companies continue to release AI models that rival the capabilities of systems developed by OpenAI and other U.S.-based AI companies.
This week, MiniMax, a startup backed by Alibaba and Tencent that has raised about $850 million in venture capital and is valued at more than $2.5 billion, launched three new models: MiniMax-Text-01, MiniMax -VL-01 and T2A. -01-HD. MiniMax-Text-01 is a text-only template, while MiniMax-VL-01 can include both images and text. The T2A-01-HD, meanwhile, generates audio, specifically speech.
MiniMax says MiniMax-Text-01, which has a size of 456 billion parameters, performs better than models like Google’s recently unveiled Gemini 2.0 Flash on benchmarks like MATH and SimpleQA, which measure a model’s ability to respond to problems math and facts. based questions. Parameters roughly correspond to a model’s problem-solving capabilities, and models with more parameters generally perform better than those with fewer parameters.
As for the MiniMax-VL-01, MiniMax says it rivals Anthropic’s Claude 3.5 Sonnet in assessments that require multimodal understanding, such as ChartQA, which tasks models with answering questions related to graphs and charts (e.g. “What is the peak value of the orange?” line in this graph?”). Of course, MiniMax-VL-01 does not outperform Gemini 2.0 Flash in many of these tests. Even OpenAI’s GPT-4o and Meta’s Llama 3.1 beat it on many.
Note that MiniMax-Text-01 has an extremely large context window. A model’s context, or context window, refers to the input (for example, text) that a model considers before generating output (additional text). With a context window of 4 million tokens, MiniMax-Text-01 can parse about 3 million words in one go, or just over five copies of “War and Peace.”
As for context (no pun intended), MiniMax-Text-01’s context window is about 31 times larger than GPT-4o and Llama 3.1.
The latest of the MiniMax models released this week, the T2A-01-HD, is an audio generator optimized for speech. The T2A-01-HD can generate a synthetic voice with adjustable cadence, pitch and tenor in approximately 17 different languages, including English and Chinese, and clone a voice from just 10 seconds of an audio recording.
MiniMax has not published benchmark results comparing the T2A-01-HD to other audio generation models. But to this reporter’s ear, the T2A-01-HD’s outputs sound on par with audio models from Meta and startups like PlayAI.
With the exception of the T2A-01-HD, which is available exclusively through the MiniMax API and Hailuo AI platform, new MiniMax models can be downloaded from GitHub and the Hugging Face AI development platform.
However, just because templates are “openly” available doesn’t mean they aren’t constrained in certain aspects. MiniMax-Text-01 and MiniMax-VL-01 are not truly open source, in the sense that MiniMax has not released the components (e.g. training data) needed to recreate them from scratch. Additionally, they are subject to MiniMax’s restrictive licensing, which prohibits developers from using the models to improve rival AI models and requires platforms with more than 100 million monthly active users to apply for a special license from MiniMax.
MiniMax was founded in 2021 by former employees of SenseTime, one of China’s largest artificial intelligence companies. The company’s projects include apps like Talkie, an AI role-playing platform along the lines of Character AI, and text-to-video models that MiniMax released in Hailuo.
Some of the MiniMax products have become the subject of minor controversies.
Talkie, which was pulled from Apple’s App Store in December for unspecified “technical” reasons, features AI avatars of public figures including Donald Trump, Taylor Swift, Elon Musk and LeBron James, none of whom appear to have agreed to be submitted to Apple’s App Store in December for unspecified “technical” reasons. app.
In December, Broadcast magazine reported that MiniMax’s video generators can play the logos of British TV channels, suggesting that MiniMax’s models were trained on content from those channels. And MiniMax is reportedly being sued by iQIYI, a Chinese video streaming service that claims MiniMax was illicitly built on iQIYI’s copyrighted recordings.
The new MiniMax models come just days after the outgoing Biden administration proposed tougher export rules and restrictions on artificial intelligence technologies for Chinese firms. Chinese companies have already been barred from purchasing advanced AI chips, but if the new rules take effect as written, companies will face tougher limits on both semiconductor technology and the models needed to launch sophisticated intelligence systems artificial.
On Wednesday, the Biden administration announced additional measures aimed at keeping sophisticated chips out of China. Chip foundries and packaging companies that want to export certain chips will be subject to broader licensing requirements unless they exercise greater oversight and due diligence to prevent their products from reaching Chinese customers.
