DeepSeek Janus Pro: Everything to know about the new AI model

DeepSeek, a Chinese AI, rocked the stock market upon releasing its R1 and V3 models. In fact, AI chip maker Nvidia and other AI companies witnessed a huge dip in their share price in the pre-open market, and they still haven't recovered as of this writing. DeepSeek has now come forward with a new MLLM (Multimodal large language model) called Janus Pro 7B, which the company claims is better than almost all of its rivals.

In this article, we will analyze the new Janus Pro 7B AI model and tell you everything you need to know.

Note: Some aspects of this article are subjective and solely reflect the writer's opinions.

---

DeepSeek Janus Pro 7B: Everything you need to know

The DeepSeek AI app on an iPhone (Image via DeepSeek)

The Janus Pro 7B is the latest Multimodal large language model from DeepSeek and is capable of simultaneously processing various types of data, including text, image, and video to generate an output.

Rowan Cheung, founder of The Rundown AI, also praised the capabilities of the new DeepSeek Janus-Pro-7B model.

What makes Janus Pro 7B special?

The Janus Pro 7B is an open-source model developed using a novel approach that allowed DeepSeek to save resources.

Here's everything that makes this model special:

• Separate pathways for visual encoding: This model can decouple the audio from the video and process them parallelly to achieve faster processing capabilities.
• Unified Transformer Architecture: This model can combine multiple types of data to deliver vastly superior output in terms of comprehension and generative capabilities.
• Built on an open-source platform: The entire model is available on the open-source platform: Hugging Face. This allows developers and researchers to analyze the model's capabilities and make contributions to it, which could vastly improve it in the long run.

---

How does it compare to other multimode AI?

The Janus Pro 7B has been benchmarked against other multimodal large language models.

Here's how it performs against popular Multimodal AI:

• MMBench: The Janus Pro 7B managed to achieve a 79.2 score in this benchmark for understanding capabilities, which is higher than others, including TokenFlow-XL (68.9 score) and MetaMorph (75.2 score).
• GenEval for Text-to-Image Generation: The Janus Pro 7B achieved an impressive 80% accuracy in this benchmark, outperforming OpenAI’s DALL-E 3 (67%) and Stable Diffusion 3 Medium (74%).

This showcases the superior capabilities of the Janus Pro 7B model when it comes to comprehension, understanding, and generative functions.

---

How did Nvidia and OpenAI respond to DeepSeek's achievement?

Nvidia acknowledged the impressive capabilities of DeepSeek and all it achieved but remained firm in its belief that its chips are crucial for AI development. Team Green also emphasized that only Nvidia can provide the kind of AI processing power that DeepSeek needs.

OpenAI's CEO Sam Altman also praised the DeepSeek AI and found it to be impressive.

---

Recent cyberattacks on DeepSeek

After the launch of the DeepSeek-R1 AI model, the company came under a barrage of cyberattacks, prompting it to limit registration in order to provide a better user experience. DeepSeek posted about it on the company's status page and noted that a fix is being implemented and that it will continuously monitor the results.