\n<\/p>\n
Chinese firms continue to release AI models that rival the capabilities of systems developed by OpenAI and other U.S.-based AI companies.<\/p>\n
This week, MiniMax<\/a>, an Alibaba- and Tencent-backed startup that has raised<\/a> around $850 million in venture capital and is valued at more than $2.5 billion, debuted<\/a> three<\/a> new models: MiniMax-Text-01, MiniMax-VL-01, and T2A-01-HD. MiniMax-Text-01 is a text-only model, while MiniMax-VL-01 can understand both images and text. T2A-01-HD, meanwhile, generates audio \u2014 specifically speech.<\/p>\n MiniMax claims that MiniMax-Text-01, which is 456 billion parameters in size, performs better than models such as Google\u2019s recently unveiled Gemini 2.0 Flash<\/a> on benchmarks like MATH and SimpleQA, which measure the ability of a model to answer math problems and fact-based questions. Parameters roughly correspond to a model\u2019s problem-solving skills, and models with more parameters generally perform better than those with fewer parameters.<\/p>\n As for MiniMax-VL-01, MiniMax says that it rivals Anthropic\u2019s Claude 3.5 Sonnet<\/a> on evaluations that require multimodal understanding, like ChartQA, which tasks models with answering graph- and diagram-related queries (e.g. \u201cWhat is the peak value of the orange line in this graph?\u201d). Granted, MiniMax-VL-01 doesn\u2019t quite best Gemini 2.0 Flash on many of these tests. OpenAI\u2019s GPT-4o<\/a> and Meta\u2019s Llama 3.1<\/a> beat it on several, as well. <\/p>\n Of note, MiniMax-Text-01 has an extremely large context window. A model\u2019s context, or context window, refers to 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 analyze around 3 million words in one go \u2014 or just over five copies of \u201cWar and Peace.\u201d<\/p>\n For context (no pun intended), MiniMax-Text-01\u2019s context window is roughly 31 times the size of GPT-4o\u2019s and Llama 3.1\u2019s.<\/p>\n The last of MiniMax\u2019s models released this week, T2A-01-HD, is an audio generator optimized for speech. T2A-01-HD can generate a synthetic voice with adjustable cadence, tone, and tenor in around 17 different languages, including English and Chinese, and clone a voice from just 10 seconds of an audio recording.<\/p>\n MiniMax didn\u2019t publish benchmark results comparing T2A-01-HD to other audio-generating models. But to this reporter\u2019s ear, T2A-01-HD\u2019s outputs sound on par with audio models from Meta<\/a> and startups like PlayAI<\/a>. <\/p>\n With the exception of T2A-01-HD, which is exclusively available through MiniMax\u2019s API and Hailuo AI platform, MiniMax\u2019s new models can be downloaded from GitHub and the AI dev platform Hugging Face. <\/p>\n Just because the models are \u201copenly\u201d available doesn\u2019t mean they aren\u2019t locked down in certain aspects, however. MiniMax-Text-01 and MiniMax-VL-01 aren\u2019t truly open source<\/a> in the sense that MiniMax hasn\u2019t released the components (e.g. training data) needed to recreate them from scratch. Moreover, they\u2019re under MiniMax\u2019s restrictive license, which prohibits developers from using the models to improve rival AI models, and requires that platforms with more than 100 million monthly active users request a special license from MiniMax. <\/p>\n MiniMax was founded in 2021 by former employees of SenseTime, one of China\u2019s largest AI firms. The company\u2019s projects include apps like Talkie, an AI-powered role-playing platform along the lines of Character AI<\/a>, and text-to-video models that MiniMax has released in Hailuo.<\/p>\n Some of MiniMax\u2019s products have become the subject of minor controversy.<\/p>\n Talkie, which was pulled from Apple\u2019s App Store in December for unspecified \u201ctechnical\u201d reasons, features AI avatars of public figures including Donald Trump, Taylor Swift, Elon Musk, and LeBron James, none of whom appear to have consented to being featured in the app.<\/p>\n In December, Broadcast magazine reported<\/a> that MiniMax\u2019s video generators can reproduce the logos of British television channels, suggesting that MiniMax\u2019s models were trained on content from those channels. And MiniMax is reportedly being sued<\/a> by iQIYI, a Chinese video streaming service that alleges MiniMax illicitly trained on iQIYI\u2019s copyrighted recordings.<\/p>\n MiniMax\u2019s new models arrive days after the outgoing Biden Administration proposed<\/a> harsher export rules and restrictions on AI technologies for Chinese ventures. Companies in China were already prevented from buying advanced AI chips, but if the new rules go into effect as written, companies will be faced with stricter caps on both the semiconductor tech and models needed to bootstrap sophisticated AI systems.<\/p>\n On Wednesday, the Biden Administration announced<\/a> additional measures focused on keeping sophisticated chips out of China. Chip foundries and packaging companies that want to export certain chips will be subjected to broader license requirements unless they exercise greater scrutiny and due diligence to prevent their products from reaching Chinese clients.<\/p>\n<\/div>\n