Alibaba launches new AI model that it says outperforms DeepSeek, China’s hottest start-up

Alibaba Group Holding released on Wednesday an upgraded version of its Qwen artificial intelligence (AI) model, which it said “comprehensively outperformed” in certain benchmark tests DeepSeek-V3, the large language model (LLM) launched in December by China’s hottest start-up.

In a statement posted on WeChat, the e-commerce giant’s cloud computing and AI arm Alibaba Cloud said its new Qwen 2.5-Max model also outperformed OpenAI’s GPT-4o and Meta Platforms’ Llama-3.1-405B in LLM performance benchmark platforms Arena-Hard and LiveBench. Alibaba owns the South China Morning Post.

The benchmark performance of Qwen 2.5-Max, part of Alibaba’s Tongyi Qianwen LLM family, was on par with Anthropic’s Claude-3.5-Sonnet model, according to Alibaba Cloud. LLMs are the technology underpinning generative AI services like ChatGPT.

Alibaba’s multimodal model is offered in various sizes, from 3 billion to 72 billion parameters, and includes both base and instruction-tuned versions. The flagship model, Qwen2.5-VL-72B-Instruct, is now accessible through the Qwen Chat platform, while the entire Qwen2.5-VL series is available on open-source platform Hugging Face and Alibaba’s own open-source community Model Scope.

Shares of Alibaba rose by almost 3 per cent in New York in recent trading, extending Tuesday’s 6.7 per cent jump, while transactions in Hong Kong were halted during the Lunar New Year holiday.