Forget Deepseek R1, apparently it is now Alibaba that is the most powerful, the cheapest, but it has the best chat bot

If you buy links to our articles, the future and its syndicate partners can earn a commission.

As Ferris observed in an ingenious way, life in the chatbot country moves pretty quickly. So forget that old news via Deepseek’s cheap and powerful LLM tank course from Nvidia. Because here comes another Chinese Tech -Riese, Alibaba, with its own new AI model that exceeds the lot. Well, it does to Alibaba.

Qwen 2.5-Max, because it is (you? You? Meet your choice), was published today according to Reuters and thus some rather bombastic claims.

“Qwen 2.5-Max exceeds … almost across the board GPT-4O, Deepseek V3 and Lama-3,1-405b,” says Alibaba. In particular, the Deepseek V3, not Deepseek R1, is the updated model, which has contributed to the fact that 600 billion US dollars will be deleted by NVIDIA’s share price in one day.

Nevertheless, these open open source models from Openai and Meta are. If the allegations of Alibaba are true, Qwen 2.5-Max is not easy.

Reuters notes that the publication of Alibaba in China is playing in a larger price war for access to AI models. It was an earlier Deepseek model that moved Alibaba to announce massive price reductions of 97% for access to its AI models.

At the moment it is unclear how resource-intensive the Qwen is 2.5-Max. What the markets really does not have to do with regard to Deepseek R1 is not the full performance, but claims that it was trained at only $ 6 million worthy of NVIDIA H800 GPUs, a small part of the costs, that are associated with the costs, with the huge GPU arrays used by Openai and Meta to train their models.

It has also turned out that the Deepseek R1 model in full precision can be carried out to only $ 6,000 PC hardware, which costs mainly through a lot of memory, but without a megabucks nvidia GPU.

The mere fact that Alibaba has a competitive model is not sovereignty. However, the questions of the hardware used and the associated costs are fascinating.

There can also be a scale to which this first chat bot technology annoys and goes together. Anyone who is familiar with Chatgpt 4o cannot help, for example, to consider the scary similarities with Deepseek R1, for example.

Your next upgrade

Best CPU to play: The top chips from Intel and AMD.
Best Gaming Motherboard: The right boards.
Best graphics card: Your perfect pixel pusher is waiting.
Best SSD to play: Get into play in front of the others.

The style and tone of the text, the delivery, the length of the answers, the actual content, the tendency to hallucination of falsehoods, the feeling of very comparable invisible waking area and due to a lack of better concept of social adaptation – everything seems to be practical for better And indeed also for the bad.

Where all of this leads is everyone’s assumption. Perhaps we remain with a load of the same, semi-use, often hallucinating bots for the foreseeable bots. Or maybe the explosion of the AGI or artificial general intelligence is right around the corner. In any case, I’m for you … no, actually I don’t welcome you. But I am fascinated by what happens next.