DeepSeek, a Chinese startup, has released an updated open-source multimodal AI model called Janus-Pro, which reportedly outperforms competitors like OpenAI's DALL-E 3 and Stable Diffusion's 3 Medium in text-to-image generation benchmarks. The new versions, Janus-Pro-1B and Janus-Pro-7B, were released alongside a research paper detailing their capabilities. This release follows the recent unveiling of DeepSeek's model R1, which demonstrated advanced reasoning abilities and was developed at a significantly lower cost compared to US rivals.
The introduction of R1 has caused significant disruption in the tech industry, raising concerns among investors due to its potential to outperform existing US models at a fraction of the cost. DeepSeek's competitive pricing strategy for accessing its models poses a challenge to the current AI business models, potentially reshaping industry assumptions.
Key takeaways:
DeepSeek has released an updated AI model called Janus-Pro, which outperforms competitors in benchmarks.
The Janus-Pro models, Janus-Pro-1B and Janus-Pro-7B, are text-to-image generators that surpass OpenAI's DALL-E 3 and Stable Diffusion's 3 Medium.
DeepSeek's previous model, R1, demonstrated advanced reasoning capabilities and was developed at a lower cost compared to US rivals.
DeepSeek is offering its models at a significantly lower price than OpenAI, potentially disrupting existing AI business models.
