Chinese AI company MiniMax has released three advanced models that it claims rival industry leaders like OpenAI and Google. Supported by Alibaba and Tencent, MiniMax has raised $850 million and is valued at over $2.5 billion.
The new models include MiniMax-Text-01, a powerful text-only model with a massive 4-million-token context window, enabling analysis of around 3 million words in one instance. It outperforms many competitors on benchmarks like MMLU and SimpleQA.
MiniMax-VL-01 is a multimodal model capable of interpreting text and images, excelling in evaluations like ChartQA, though it falls short against some rivals, including OpenAI’s GPT-4o.
Meanwhile, T2A-01-HD is an audio generator that produces lifelike speech in 17 languages and clones voices using just 10 seconds of recording.
Despite being available on GitHub and Hugging Face, MiniMax’s models are not fully open source. Restrictions prevent their use in improving competing AI models, and platforms with over 100 million users must obtain special licenses.
Founded in 2021 by ex-SenseTime employees, MiniMax also developed Talkie, an AI-powered role-playing app featuring avatars of public figures like Donald Trump and Taylor Swift, which faced removal from Apple’s App Store.
Controversies have arisen over allegations of unauthorized use of copyrighted materials, including logos and recordings.
The launch of these models coincides with the Biden administration’s proposed stricter export rules targeting AI technologies and chips for Chinese firms.
These measures aim to limit access to resources critical for developing advanced AI systems, potentially impacting the growth of companies like MiniMax.
