DeepSeek unleashes Janus-Pro 7B model: focuses on task-specific AI models, rivals DALL-E 3

DeepSeek's introduction into the world of AI assistants has shaken the world, seeing NVIDIA stock bleeding out $500 billion with no signs of stopping... and now the Chinese AI startup has unleashed its new multimodal AI model: Janus-Pro 7B.

2

Janus-Pro 7B is under an MIT license meaning it can be used commercially without any restrictions, with DeepSeek explaining Janus-Pro as a "novel autoregressive framework" capable of analyzing and creating images. The company says that on two AI evaluation benchmarks -- GenEval and DPG-Bench -- the largest Janus-Pro model, Janus-Pro-7B, beats DALL-E 3 as well as other AI models including PixArt-alpha, Emu3-Gen, and Stability AI's Stable Diffusion XL.

DeepSeek explained on Hugging Face: "Janus-Pro surpasses previous unified model and matches or exceeds the performance of task-specific models. The simplicity, high flexibility, and effectiveness of Janus-Pro make it a strong candidate for next-generation unified multimodal models".

Coming seemingly out of nowhere, the Chinese AI company has risen to the top of the Apple App Store charts, beating out OpenAI's dominant ChatGPT as the most popular free app on the App Store. DeepSeek's AI model is far, far cheaper to train than competing AI models, costing just $5.6 million to train on far less powerful (and far cheaper expensive) hardware.

• Read more: NVIDIA shares bleed $384B in a few hours after DeepSeek shocks AI world
• Read more: DeepSeek's new AI app: top free app on the Apple App Store, ahead of ChatGPT
• Read more: Chinese AI firm DeepSeek has 50,000 NVIDIA H100 AI GPUs says CEO, even with US restrictions

AI competitors require hundreds of millions of dollars to train their AI models, as well as hundreds of billions of dollars of AI hardware... mostly from NVIDIA, which has seen Team Green bleeding out over $500 billion from its market cap in the last 24 hours alone.