DeepSeek, the viral AI company, has unveiled a new set of multimodal AI models called Janus-Pro, which claim to outperform OpenAI’s DALL-E 3. Available on Hugging Face, the models range from 1 billion to 7 billion parameters, offering a compact yet powerful alternative to larger models.
Key Features of Janus-Pro
- Size and Performance: Janus-Pro models boast high efficiency, with the largest, Janus-Pro-7B, outperforming DALL-E 3 and other popular models like Stable Diffusion XL on benchmarks like GenEval and DPG-Bench.
- Flexibility: Built under an MIT license, the models can be used commercially without restrictions.
- Multimodal Capabilities: Janus-Pro can analyze and generate images, albeit with a maximum resolution of 384 x 384 for smaller models.
Industry Disruption
DeepSeek’s post on Hugging Face highlights the models’ capabilities:
“Janus-Pro surpasses previous unified models and matches or exceeds task-specific performance, making it ideal for next-generation AI solutions.”
The company, backed by High-Flyer Capital Management, has gained widespread attention after its chatbot app topped the Apple App Store charts. Many analysts now question whether the U.S. can retain its lead in AI as DeepSeek continues to innovate with cost-effective and powerful models.
