DeepSeek just released an updated open-source multimodal AI model.
The Chinese startup on Monday shared a research paper and released updated versions of the model, called Janus-Pro-1B, and Janus-Pro-7B.
According to its paper, DeepSeek says Janus-Pro outperforms OpenAI's DALL-E 3 and Stable Diffusion's 3 Medium text-to-image generators across multiple benchmarks.
The latest release comes after DeepSeek unveiled its model R1 last week, which showed new "reasoning" capabilities.
R1's release sent shockwaves across the tech industry and spooked investors. That's because DeepSeek demonstrated a breakthrough AI model that could outperform US rivals, but at what it says is just a fraction of the cost.
DeepSeek is pricing access to its models way under what OpenAI charges, potentially undermining AI business models and assumptions across the sector.
Contact the reporter Jyoti Mann via email at jmann@jkmperu.com or via Signal at jyotimann.11. Reach out via a nonwork device.
