Chinese AI company MiniMax releases new models it claims are competitive with the industry’s best

Chinese firms continue to release AI models that rival the capabilities of systems developed by OpenAI and other US-based AI firms.

this week, minimaxAn Alibaba- and Tencent-backed startup that has Raised About $850 million in venture capital and over $2.5 billion in value, self expression three New models: MiniMax-Text-01, MiniMax-VL-01, and T2A-01-HD. The MiniMax-Text-01 is a text-only model, while the MiniMax-VL-01 can understand both images and text. The T2A-01-HD, meanwhile, produces audio — specifically speech.

Minimax claims that minimax-text-01, which has a size of 456 billion parameters, outperforms Google’s recently exposed models. Gemini 2.0 Flash In benchmarks such as MATH and SimpleQA, which measure the model’s ability to answer math problems and data-based questions. Parameters roughly correspond to a model’s problem-solving ability, and models with more parameters generally perform better than models with fewer parameters.

As for the MiniMax-VL-01, MiniMax says it rivals Anthropic’s Claude 3.5 Sonnet Assessments that require multimodal understanding, such as ChartQA, which make models work by answering graph- and diagram-related questions (eg, “What is the maximum value of the orange line on this graph?”). Well, the MiniMax-VL-01 isn’t very good for the Gemini 2.0 flash in many of these tests. of OpenAI GPT-4o and meta’s Llama 3.1 Quite a few beat it as well.

Note that MiniMax-Text-01 has a very large context window. A model’s context, or context window, refers to the input (eg, text) that a model considers before generating output (additional text). With a context window of 4 million tokens, MiniMax-Text-01 can analyze about 3 million words simultaneously – or just over five copies of “War and Peace.”

For context (no pun intended), MiniMax-Text-01’s context window is about 31 times the size of GPT-4o and Llama 3.1.

MiniMax’s latest model 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, tone and tenor in about 17 different languages, including English and Chinese, and clone a voice from just 10 seconds of an audio recording.

The MiniMax T2A-01-HD did not reveal benchmark results compared to other audio-producing models. But to this reporter’s ears, the T2A-01-HD’s outputs sound on par with audio models meta And like startups Play AI.

Except for the T2A-01-HD, which is exclusively available through MiniMax’s API and the Hailuo AI platform, MiniMax’s new models can be downloaded from GitHub and the AI ​​dev platform Hugging Face.

Just because models are “openly” available doesn’t mean they aren’t locked in certain aspects, however. MiniMax-Text-01 and MiniMax-VL-01 Not really open source In the sense that Minimax did not expose the necessary components (eg, training data) to be recreated from scratch. Furthermore, they are under a restrictive license from MiniMax, which prohibits developers from using the models to improve competing AI models, and platforms with more than 100 million monthly active users request an exclusive license from MiniMax.

MiniMax was founded in 2021 by former employees of SenseTime, one of the largest AI companies in China. The company’s projects include apps like Talkie, an AI-powered role-playing platform Character AIAnd the text-to-video model that MiniMax released on Hailuo.

Some of MiniMax’s products have been the subject of minor controversy.

The talkie, pulled from Apple’s App Store in December for unspecified “technical” reasons, featured AI avatars of public figures including Donald Trump, Taylor Swift, Elon Musk and LeBron James, none of whom appeared to have consented to being featured. The app

Broadcast Magazine in December Report The fact that Minimax’s video generators can reproduce British television channel logos suggests that Minimax’s models were trained on the content of those channels. and is said to be minimax A case is being filed by iQiyi, a Chinese video streaming service that alleges that MiniMax is illegally trained on iQiyi’s copyrighted recordings.

The new Minimax models come days after the outgoing Biden administration Recommended Strict export rules and restrictions on AI technology for Chinese enterprises. Chinese companies were already barred from buying advanced AI chips, but if the new rules go into effect as written, companies will face strict caps on both the semiconductor technology and models needed to bootstrap sophisticated AI systems.

Biden administration on Wednesday announcement Additional measures focused on keeping sophisticated chips out of China. Chip foundries and packaging companies that want to export certain chips will be subject to greater licensing requirements unless they do greater vetting and due diligence to prevent their products from reaching Chinese clients.