Chinese firms proceed to release AI models that rival the capabilities of systems developed by OpenAI and other U.S.-based AI corporations.
This week, MiniMaxan Alibaba and Tencent-backed startup that educated Approximately $850 million in enterprise capital and valued at greater than $2.5 billion. debuted three latest 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 each images and text. T2A-01-HD, then again, produces audio – especially speech.
MiniMax claims that MiniMax-Text-01, which is 456 billion parameters in size, performs higher on benchmarks like MMLU and SimpleQA, which measure a model's ability to reply math problems and facts, than models like Google's recently unveiled Gemini 2.0 Flash. based questions. Parameters roughly correspond to a model's problem-solving capabilities, and models with more parameters generally perform higher than those with fewer parameters.
As for MiniMax-VL-01, MiniMax says it competes with Anthropic's Claude 3.5 Sonnet in evaluations that require multimodal understanding, similar to ChartQA, which tasks models with answering graph- and chart-related questions (e.g. ” What is the height value of the orange line on this graph?”). Admittedly, MiniMax-VL-01 doesn't quite outperform Gemini 2.0 Flash in lots of these tests. OpenAI's GPT-4o and an open model called InternVL2.5 also beat it on several counts.
What's notable is that MiniMax-Text-01 has an especially large context window. A model's context or context window refers to input (e.g. 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 directly – or simply over five copies of War and Peace.
As for context (no pun intended): MiniMax-Text-01's context window is roughly 31 times larger than that of GPT-4o and Llama 3.1.
The latest MiniMax model released this week, T2A-01-HD, is an audio generator optimized for speech. T2A-01-HD can generate an artificial voice with adjustable cadence, timbre and tenor in around 17 different languages, including English and Chinese, and clone a voice from just 10 seconds of an audio recording.
MiniMax has not released any benchmark results comparing the T2A-01-HD to other audio generation models. But to this reporter's ears, the T2A-01-HD's outputs sound on par with Meta and startups like PlayAI.
Except for T2A-01-HD, which is exclusively available through MiniMax's API and Hailuo AI platform, MiniMax's latest models might be downloaded from GitHub and the Hugging Face AI development platform.
However, simply because the models can be found “openly” doesn’t mean that they will not be locked in certain elements. MiniMax-Text-01 and MiniMax-VL-01 will not be truly open source within the sense that MiniMax has not released the components (e.g. training data) required to construct them from scratch. Additionally, they’re subject to MiniMax's restrictive license, which prohibits developers from using the models to enhance competing AI models and requires platforms with greater than 100 million monthly lively users to request a special license from MiniMax.
MiniMax was founded in 2021 by former employees of SenseTime, certainly one of China's largest AI corporations. The company's projects include apps similar to Talkie, an AI-powered role-playing platform modeled on character AI, and text-to-video models that MiniMax released in Hailuo.
Some of MiniMax's products have change into the topic of minor controversy.
Talkie, which was faraway 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 seem like theirs The app has agreed to be published.
In December, Broadcast Magazine reported that MiniMax's video generators can reproduce the logos of British television channels, suggesting that MiniMax's models were trained on content from these channels. And MiniMax ought to be too is sued by iQiyi, a Chinese video streaming service, which accuses MiniMax of illegally working on iQiyi's copyrighted recordings.
MiniMax's latest models come days after the outgoing Biden administration proposed stricter export rules and restrictions on AI technologies for Chinese corporations. Companies in China were already blocked from purchasing advanced AI chips, but when the brand new rules take effect as written, corporations will face stricter caps on each the semiconductor technology and the models needed to launch advanced AI systems are.
On Wednesday, the Biden administration announced Additional measures focused on keeping advanced chips out of China. Chip foundries and packaging corporations that wish to export certain chips are subject to broader licensing requirements unless they apply greater control and due diligence to stop their products from reaching Chinese customers.
