DeepSeek, the Chinese startup that has managed to make a mockery of Silicon Valley’s capital-bloated AI oligarchy, has done it again. On Monday morning, the company announced the release of yet another open-source AI system, this one an image generator that—the company claimed—could best OpenAI’s DALL-E and Stability AI’s Stable Diffusion generators. The model—Janus-Pro-7B—was announced in a technical paper shared on DeepSeek’s GitHub page Monday.
It is an update of Janus, a simpler model that was released last October. Janus promises to bring multi-modal automation to new heights, and can conduct both image generation and image analysis, the company said. The paper claims that DeepSeek’s model outperforms both DALL-E and Stable Diffusion on multiple AI benchmarks.
The new model release follows news that the startup’s R1 model, which was released in December, has surpassed ChatGPT in downloads on the Apple app store . That would be news enough on its own, but DeepSeek also claims it was able to create R1—which is an open-source “reasoning” model—in only two months with, what it claims, was only $6 million. Given that the U.
S.’s top AI company, OpenAI, is worth over a hundred billion dollars and is closed-source, you can see why America’s tech oligarchs might be freaking out about this. On the same morning that DeepSeek released its image-generator, the company said it was suffering from “large-scale malicious attacks” on its network, Reuter.
