DeepSeek, the viral AI company, has released a new set of multimedia AI models that it claims can outperform OpenAI’s DALL-E 3.
Models, which are Available for download From the Hugging Face AI development platform, part of a new model family that DeepSeek calls Janus Pro. They range in size from 1 billion parameters to 7 billion parameters. The parameters roughly correspond to the model’s problem-solving skills, and models with more parameters generally perform better than those with fewer.
Janus Pro is under a license from the Massachusetts Institute of Technology (MIT), which means it can be used commercially without restrictions.
Janus Pro, which DeepSeek describes as a “new regression framework,” can analyze and create new images. According to the company, in two AI benchmarks, GenEval and DPG-Bench, the largest Janus Pro model, the Janus Pro 7B, outperforms the DALL-E 3 as well as models such as PixArt-alpha, Emu3-Gen, and Ability AI’s Stable. Spread XL.
Some of these models are on the older side. And Janus Pro can only analyze and create small images – images with a resolution of 384 x 384. But the performance of the Janus Pro family is impressive, considering the small sizes of the models.
“Janus Pro outperforms the previous unified model and matches or exceeds the performance of mission-specific models,” DeepSeek he writes in a post on Hugging Face. “The simplicity, high flexibility and effectiveness of Janus Pro make it a strong candidate for the next generation of unified multimedia models.”
DeepSeek, a Chinese artificial intelligence lab largely funded by quantitative trading firm High-Flyer Capital Management, burst into the mainstream consciousness this week after… His chatbot app rose to the top of Apple’s App Store charts. DeepSeek’s language models, trained using powerful computational techniques, have led many analysts — and Wall Street technologists — to question whether the United States can maintain its lead in the AI race, and whether demand for AI chips will continue.
