DeepSeek, a Chinese startup, has just unveiled its latest AI model, Janus-Pro, which focuses on generating high-quality images from text prompts.
The model is designed to compete directly with US-based models like DALL-E 3 and Stable Diffusion, offering improved performance in areas such as image accuracy and quality. This new release follows the success of DeepSeek’s R1 model, which gained attention for its fast and logical responses, built at a fraction of the cost compared to US models.
Despite using less advanced Nvidia chips, DeepSeek has raised questions about how it competes with the US, especially given the absence of cutting-edge technology. The company’s recent app success, outpacing ChatGPT on the Apple App Store, is further proof of its rising influence in AI development.
The launch of Janus-Pro reflects China’s ambition to establish a strong presence in the global AI race, with the timing coinciding with increased scrutiny of Chinese tech companies, including concerns over data privacy, as seen with TikTok.
Last week, President Donald Trump announced a new AI infrastructure initiative, pledging up to $500 million in partnership with OpenAI and other firms. Janus-Pro is available for download on the AI developer platform Hugging Face.
In the model’s introduction on the download page, DeepSeek emphasized that Janus-Pro outperforms its predecessor and matches or exceeds the performance of task-specific models. The flexibility and simplicity of the model make it a promising candidate for next-generation AI tools.
Janus-Pro’s design features a range of model sizes, from 1 billion to 7 billion parameters, significantly impacting its ability to solve complex problems. DeepSeek also described Janus-Pro as a “novel autoregressive framework” that resolves previous issues by decoupling the steps for analyzing and generating images while maintaining a unified system.
