Chinese companies continue to release AI models that rival the capabilities of systems developed by OpenAI and other US-based AI companies. This week, MiniMax, an Alibaba- and Tencent-backed startup that has raised about $850 million in venture capital and is valued at more than $2.5 billion, debuted 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, reproduces audio – specifically speech. MiniMax claims that MiniMax-Text-01, whose size is 456 billion, performs better than models such as Gemini 2.0 Flash that Google recently launched in benchmarks like MATH and SimpleQA, which measure the ability of models to answer mathematical problems and fact-based questions. Parameters are roughly proportional to the model’s problem-solving skill, and models with more parameters generally perform better than models with fewer parameters. As for the MiniMax-VL-01, MiniMax says it competes with Anthropic’s Claude 3.5 Sonnet in evaluations that require multimodal understanding, like ChartQA, which tasks the model with answering questions related to graphs and diagrams (eg “What is the value of the orange peak? line in this graph?”). Of course, the MiniMax-VL-01 does not best the Gemini 2.0 Flash in many of these tests. OpenAI’s GPT-4o and Meta’s Llama 3.1 also beat some. Note, MiniMax-Text-01 has a very large context window. The model context, or context window, shows the input (eg, text) that the model considers before producing the output (additional text). With a context window of 4 million tokens, MiniMax-Text-01 can analyze about 3 million words at a time – or just over five copies of “War and Peace.” For context (useless), the MiniMax-Text-01 context window is about 31 times the size of 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 synthetic sounds with adjustable rhythm, pitch, and tenor in about 17 different languages, including English and Chinese, and clone sounds from just 10 seconds of audio recording. MiniMax does not publish benchmark results comparing the T2A-01-HD to other audio reproduction models. But to this reporter’s ears, the T2A-01-HD produces sound that matches audio models from Meta and startups like PlayAI. Except for the T2A-01-HD, which is exclusively available through the MiniMax API and the Hailuo AI platform, the new MiniMax models can be downloaded from GitHub and the AI dev platform Hugging Face. Just because the models are available “open” does not mean that they are not locked in certain aspects, however. MiniMax-Text-01 and MiniMax-VL-01 are not open source because MiniMax has not yet released the components (e.g. training data) needed to recreate them from scratch. Additionally, they are under MiniMax’s restrictive license, which prohibits developers from using the model to improve competing AI models, and requires platforms with more than 100 million monthly active users to request a special 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 applications like Talkie, an AI-powered role-playing platform along the lines of Character AI, and the text-to-video model that MiniMax has released on Hailuo. Some MiniMax products have been the subject of a little controversy. Talkie, which was pulled from the Apple 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 in the app. In December, Broadcast magazine reported that the MiniMax video generator was able to reproduce the logo of a British television channel, suggesting that the MiniMax model was trained on content from the channel. And MiniMax is reportedly being sued by iQIYI, a Chinese video streaming service that claims MiniMax illegally trained on iQIYI’s copyrighted footage. The new MiniMax model comes days after the outgoing Biden Administration proposed tougher export rules and restrictions on AI technology for Chinese ventures. Companies in China have been prevented from buying advanced AI chips, but if the new rules go into effect as written, companies will face tighter caps on the semiconductor technology and models needed to bootstrap advanced AI systems. On Wednesday, the Biden Administration announced additional measures focused on keeping advanced chips from China. Chip foundries and packaging companies that want to export certain chips will be subject to more licensing requirements in addition to greater oversight and due diligence to prevent their products from reaching Chinese clients.
China’s AI company MiniMax released a new model it claims is competitive with the best in the industry
