The Chinese Deepseek attracted attention with the fact that advanced artificial intelligence (AI) models became popular. The company spent only less than $ 6 million to train AI models. In contrast, the amount for OpenAI’s DALL-E 3 model was around $ 100 million.

Deepseek’s success has just gone beyond being a number in the App Store; We are talking about a development that even loses 400 billion dollars in the US market in the US market. As such, it was inevitable that the service was attacked by cyber attacks and capacity problems.

Janus-Pro’s groundbreaking success

Deepseek, the Chinese-based AI laboratory, who is trying to overcome these difficulties, also announced the AI ​​model from an open-source text to the Janus-pro. This new model arouses a huge repercussion and performs better in a few criteria from OpenAI’s DALL-E 3, Stability AI’s Stable Diffusion and other similar models.

• China’s artificial intelligence Deepseek, how did it suddenly become so popular?

Janus-Pro is the updated version of the Janus model, which was released in the end of last year. Janus-pro is offered in different sizes; These include options ranging from 1 billion parameters to 7 billion parameters. According to the data shared by Deepseek, Janus-Pr-7B, the largest model, has excellent performance in image production and analysis, leaving its rivals Pixart-Alpha, EMU3-Gen and SDXL in industrial standards such as Geneval and DPG-Bench. The Janus-Pro-7B can be downloaded free of charge via the HuggingfaceaI platform, which is very popular in the field of machine learning.

New Approach: Visual Coder and Flexibility

The Janus-Pro-7B is based on an autoregressive frame that distinguishes visual coding processes using a combined transformer architecture. This approach not only alleviates the conflict between the role of production and understanding of the visual coder, but also increases the flexibility of the model. Compared to models designed specifically for singular processes, Janus-Pro leaves behind its competitors in multiple tasks. However, it does not remain in the shadow of the performance offered by special purposes.

Competitive difficulty and application areas

The release of Janus-Pro, after Deepseek’s previous success, caused a great competition with the effect of the new R1 language model, which offers similar features to GPT-4. The low cost of these advanced models has created a shock effect in the US AI industry. Such new models promise a great transformation at more affordable prices compared to traditional AI applications in the sector.