Alibaba has released its latest AI model, Qwen 2.5-Max, which it claims to surpass the ‘leading state-of-the-art models’, including DeepSeek V3.

Alibaba’s cloud unit was cited by Reuters as saying in an announcement posted on its official WeChat account as saying: “Qwen 2.5-Max outperforms … almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B.”

Alibaba said its Qwen2.5-Max outperforms DeepSeek V3 in various benchmarks, including Arena-Hard, LiveBench, LiveCodeBench, and GPQA-Diamond. It also showed strong performance in other assessments such as MMLU-Pro.

The evaluation of Alibaba’s AI model included comparisons with other leading models, such as the open-weight MoE model DeepSeek V3, the dense model Llama-3.1-405B, and Qwen2.5-72B.

In a blog, Alibaba’s Qwen said: “Now Qwen2.5-Max is available in Qwen Chat, and you can directly chat with the model, or play with artifacts, search, etc.”

Recently, DeepSeek released AI assistant and R1 model which led to a significant drop in tech shares in Silicon Valley.

The low development and operational costs of the Chinese startup’s models have raised questions about the large expenditures of US-based AI firms.

ByteDance, the parent company of TikTok, also entered the fray with an update to its AI model, claiming superiority over OpenAI’s offerings in benchmark tests.

This move mirrored DeepSeek’s assertions of its R1 model’s competitive performance.

The release of DeepSeek-V2 in May 2024, initiated price competition among AI models in China.