Alibaba unveiled Qwen 2.5-Max and claimed it outperformed DeepSeek-V3, GPT-4o and Llama-3.1-405B
The launch came just days after Chinese AI startup DeepSeek rolled out a free AI assistant, which it claims uses less data
While DeepSeek-V3 model was trained using NVIDIA H800 chips at a cost of $6 Mn, OpenAI’s GPT-4 model was trained using 100K NVIDIA GPUs at a cost of around $100 Mn
Chinese tech giant Alibaba has rolled out a new version of its Qwen 2.5 artificial intelligence (AI) model, claiming that it outperformed DeepSeek’s most advanced open source AI model that has taken the world by storm.
The company’s cloud computing unit Alibaba Cloud said in its official WeChat account that Qwen 2.5-Max surpassed OpenAI’s GPT-4o, DeepSeek-V3 and Meta’s Llama-3.1-405B ‘almost’ across the board, Reuters reported.
The move comes at a time when Chinese AI startup DeepSeek has shattered beliefs about the cost of AI, putting immense pressure on leading AI firms like OpenAI and Meta.
Last week, DeepSeek launched its free AI assistant DeepSeek-V3, claiming that it uses less data and costs only a fraction of services offered by incumbents.
DeepSeek researchers wrote in a paper last month that the DeepSeek-V3 model, released on January 10, used NVIDIA’s H800 GPUs for training, and cost less than $6 Mn. In comparison, OpenAI’s GPT-4 model was trained using 100K NVIDIA GPUs (the more advanced H100) at the cost of $100 Mn.
The fact that DeepSeek APIs cost up to 95% less than OpenAI’s latest models prompted investors to question the huge spending by AI behemoths in the US.
The emergence of the low-cost Chinese AI model and its rising popularity on app stores worried investors, setting a selloff in tech stocks such as NVIDIA, Broadcom, Alphabet and Microsoft earlier this week.
DeepSeek has also released an AI model called Janus-Pro-7B for AI-led image generation, which it claims outperforms OpenAI’s DALL-E 3 and Stability AI’s Stable Difussion. This is likely to put further strain on tech stocks.
However, for Indian startups looking to build their own large language models from scratch, DeepSeek’s meteoric rise is good news. While founders of AI startups in India have thus far lamented about the need of intensive capital to build LLMs, DeepSeek has shattered these beliefs.
Recently, AI search engine Perplexity’s founder and CEO Aravind Srinivas said that India can build cheaper AI models through ingenuity and strategic investments in open source models.
“I think that’s possible for AI (to train models frugally), given the recent achievements of DeepSeek. So, I hope India changes its stance from wanting to reuse models from open-source and instead trying to build muscle to train their models that are not just good for Indic languages but are globally competitive on all benchmarks,” Srinivas said.
