DeepSeek, a Chinese company that has recently gained popularity and sparked concerns among Western AI companies, has released a new set of multimodal AI models that DeepSeek claims can outperform OpenAI’s DALL-E 3 image generator.
DeepSeek, a Chinese company that has recently gained popularity and sparked concerns among Western AI companies, has released a new set of multimodal AI models that DeepSeek claims can outperform OpenAI’s DALL-E 3 image generator.
The models, which are available for download from the Hugging Face AI developer platform, are part of a new family of models that DeepSeek calls Janus Pro. They have between 1 billion and 7 billion parameters, a measure that correlates with a model’s skill at solving problems. Models with more parameters tend to perform better than models with fewer parameters, TechCrunch reports.
Janus Pro, which DeepSeek describes as a «new autoregression framework,» can both analyze and create new images. According to the company, in two AI benchmarks, GenEval and DPG-Bench, the largest Janus Pro model, Janus Pro 7B, outperforms DALL-E 3, as well as models like PixArt-alpha, Emu3-Gen, and Stability AI’s Stable Diffusion XL.
Of course, some of these models are older, and Janus Pro can only analyze and generate small images with resolutions up to 384×384. But the performance of Janus Pro is impressive considering the compact size of the models.
«Janus Pro outperforms the previous unified model and matches or exceeds the performance of models for specific tasks. Janus Pro’s simplicity, high flexibility, and efficiency make it a strong candidate for the role of the next-generation unified multimodal model», — DeepSeek writes in a post on Hugging Face.
Janus Pro is licensed by the Massachusetts Institute of Technology, meaning this AI can be used for commercial purposes without restrictions.
DeepSeek, a Chinese artificial intelligence lab funded largely by High-Flyer Capital Management, recently gained popularity after its chatbot rose to the top of the Apple App Store charts.
DeepSeek’s language models, which were trained using efficient computational methods, have many analysts and technologists in the West wondering whether the US can maintain its lead in the AI race and whether demand for AI chips will persist. The worry has sent tech stocks, including top graphics processor maker Nvidia, tumbling, and Mark Zuckerberg hastening to announce that Meta plans to invest $60 billion in AI by 2025.
The hype around DeepSeek became so great that even monobank co-founder Oleg Gorohovsky and the head of the Ministry of Digital Affairs, Mykhailo Fedorov, wrote cryptic, laconic posts about the startup. The latter later published a more detailed post on Telegram.
«We think DeepSeek is more of an evolution than a revolution: they have successfully combined existing developments and done it cheaper. But this is unlikely to affect the race to create AGI (super-AI), which remains the main goal of the industry. The current market reaction may be exaggerated, but investors are starting to think about the effectiveness of large expenditures in AI (remember 500 billion on Stargate)», — Fedorov noted.
Previously, dev.ua did a detailed analysis of how DeepSeek managed to outperform its competitors.
