DeepSeek’s AI model sets new benchmark in image generation

A breakthrough by Chinese startup DeepSeek has shaken the AI sector, with its Janus Pro model surpassing rivals in image generation benchmarks.
DeepSeek's impact on future electricity demand cannot be solely determined, according to Japan's METI.

Chinese AI startup DeepSeek has announced that its Janus-Pro-7B model has surpassed competitors, including OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion, in benchmark rankings for text-to-image generation. This achievement solidifies DeepSeek’s reputation as a key player in the rapidly evolving AI market.

According to a technical report, the Janus-Pro model builds upon its predecessor by incorporating enhanced training processes, higher-quality data, and advanced scaling, resulting in improved stability and more detailed image outputs. The company credited the inclusion of 72 million high-quality synthetic images, combined with real-world data, for the model’s superior performance.

This success follows the launch of DeepSeek’s new AI assistant based on the DeepSeek-V3 model, which has become the top-rated free app in the US Apple App Store. The news sent shockwaves through the tech industry, leading to declines in shares of companies like Nvidia and Oracle, as investors reassessed the competitive dynamics in AI development.

OpenAI and Stability AI have yet to comment on the claims. DeepSeek’s achievements highlight the growing influence of Chinese firms in cutting-edge AI innovation, setting the stage for heightened competition in the global tech market.