After DeepSeek, Alibaba’s new AI model Qwen claims edge over ChatGPT, Llama

China is doubling down on the AI race with back-to-back breakthroughs. Just as newcomer DeepSeek shook up the U.S. AI landscape, e-commerce giant Alibaba has unveiled its newest AI model Qwen 2.5 Max, and it claims it outperforms DeepSeek’s R1 model and even the likes of OpenAI’s GPT-4o and Meta’s Llama.

The company compares the performance of Qwen2.5-Max with leading state-of-the-art models, including DeepSeek V3, GPT-4o, and Claude-3.5-Sonnet. "Qwen 2.5-Max outperforms ... almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B," says Alibaba's Cloud via its WeChat account, global news agency Reuters reported.

Alibaba Cloud's latest blog claims it beats them on parameters like Arena-Hard, LiveBench, LiveCodeBench, and performs well across other benchmarks like MMLU-Pro, and GPQA-Diamond. These AI benchmarks evaluate large language models (LLMs) across capabilities like reasoning, knowledge retrieval, problem-solving, and coding.

"Qwen2.5-Max outperforms DeepSeek V3 in benchmarks such as Arena-Hard, LiveBench, LiveCodeBench, and GPQA-Diamond, while also demonstrating competitive results in other assessments, including MMLU-Pro," says Alibaba Cloud in a blog post.

Alibaba says its base models have shown "significant advantages" across most benchmarks, and is optimistic that advancements in post-training techniques will elevate the next version of Qwen2.5-Max to new heights. "When comparing base models, we are unable to access the proprietary models such as GPT-4o and Claude-3.5-Sonnet. Therefore, we evaluate Qwen2.5-Max against DeepSeek V3, a leading open-weight MoE model, Llama-3.1-405B, the largest open-weight dense model, and Qwen2.5-72B, which is also among the top open-weight dense models."

The company claims the results of this comparison show Qwen 2.5 Max "outperforming" DeepSeek-V3 and Llama3.1-405B on several benchmarks. Notably, DeepSeek's two AI models, DeepSeek V3 and R1, which were released recently, not only took the US AI community by storm but also wiped out $1 trillion from the top chip manufacturers' market value in the US as shares of Nvidia, Broadcom, Meta, Microsoft, Oracle, among others, saw major fall.

DeepSeek's AI model performs on par with several of the advanced AI models by American companies, and the surprising part is its cost. Unlike OpenAI or Google's AI models, which cost around $100 million to train, DeepSeek's latest model cost just $5.6 million. VC and adviser to US President Donald Trump Marc Andreessen has called the development a "This is AI's Sputnik moment", and "one of the most amazing and impressive breakthroughs I’ve ever seen".

What is DeepSeek?

DeepSeek (DS) is 100% owned by a highly successful AI-driven quant fund in China, High-Flyer. High-Flyer created DS in April 2023 to focus on artificial general intelligence (AGI/LLM). V2 was launched in May 2024, with a reported cost of only Rmb2/output token. It achieved a No. 7 ranking on the University of Waterloo's LLM leaderboard. Last month, it launched V4, trained on a data set of 14.8tn tokens (13tn for GPT4-o), at a training cost of only $5.6 million. That is less than 10% of the cost of Meta's Llama.

DeepSeek also indicated V3's performance exceeded that of Llama 3.1 and Qwen 2.5, while matching GPT4-o and Claude 3.5 Sonnet. "DS's architecture is based on a Mixture of Experts (MoE) and Multi-head Latent Attention (MLA). Each MoE model has 200bn data parameters, and each query would activate only 20bn parameters, which lowers the inferencing cost and shortens the response time. It is an open-source model, available at Hugging Face," says global brokerage Jefferies in its latest analysis.