China’s Deepseeker1 has been unveiled, its benchmark score is at the top, and the price is about one-fifth cheaper than the latest ChatGPT model, shocking.

Learning is a very small budget of $5.5 million, which is as low as a monthly salary for a big tech organization leader

Especially for learning, I was shocked to learn that the hardware reportedly borrowed 2 Nvidia H800s for 2 months for 2 dollars per hour..

Big tech is also in turmoil.
It all started with Deepseek V3, a model that has already exceeded Llama 4 in benchmark testing

To make matters worse, an “unidentified Chinese company” trained on a “small training budget of $5.5 million” has emerged, making the situation worse.

Engineers are going crazy to dissect Deepseek and copy everything possible from him. I’m really not exaggerating.

Management is struggling with how to justify the huge costs of GenAI organizations.

How can management face leadership when the “leaders” of the GenAI organization are paid more than the overall training costs of Deepseek V3 and there are dozens of such “leaders”?

Deepseek R1 has made the situation even more serious. It’s confidential, so I can’t reveal it in detail, but it will be released soon.

🐯Thinking of a tiger🐯

• Even if you try the free version of deepseek, the performance is good, and the price of using an API is overwhelming at 5% compared to ChatGPT
• Scratch search sometimes I’m surprised to get data from over 50 sites and answer it..
• When I asked right after the presentation, I said ChatGPT, and there are information security issues, but China’s development is scary..
• Most of the team is made up of graduates from Chinese internal universities and 5 years of PhD. Isn’t it the fear of the engineering number..