The artificial intelligence company DeepSeek has released a new set of multimodal AI models, which, as it claims, can outperform DALL-E 3 from OpenAI.
The models, available for download from the AI developer platform Hugging Face, are part of a new family called Janus Pro. They range from 1 billion to 7 billion parameters. Models with a higher number of parameters generally perform better than those with fewer. Janus Pro is published under the Massachusetts Institute of Technology (MIT) license, meaning AI can be used for commercial purposes without restrictions.
DeepSeek describes Janus Pro as “a new autoregression system.” The models can analyze images or create new ones. According to the company’s data, on two AI evaluation tests, GenEval and DPG-Bench, the largest Janus Pro model, Janus Pro 7B, beats DALL-E 3, as well as the PixArt-alpha, Emu3-Gen, and Stable Diffusion XL models.
Of course, some of these models are older, and Janus Pro can only analyze and create small images with a resolution up to 384 x 384. But the performance of the Janus Pro family impresses, considering the compact size of the models.
“Janus Pro surpasses the previous unified model and matches or exceeds the performance of task-specific models. The simplicity, high flexibility, and efficiency of Janus Pro make it a strong candidate from the next generation of unified multimodal models,” writes DeepSeek on Hugging Face.
The Chinese artificial intelligence laboratory DeepSeek is primarily funded by the trading company High-Flyer Capital Management, this week broke into the public consciousness with a chatbot application that soared to the top of the Apple App Store charts. The language models of DeepSeek, which were trained using computationally efficient methods, made many Wall Street analysts and tech experts doubt whether the US can maintain its leadership in the AI race.
Source: TechCrunch
Spelling error report
The following text will be sent to our editors: