Jan 30, 2025 16:00:00

Chinese IT giant Alibaba releases 'Qwen2.5-Max', an AI model with higher performance than GPT-4o and DeepSeek-V3

Qwen , an AI research team at Alibaba Cloud, released the AI model ' Qwen2.5-Max ' on January 28, 2025. Qwen2.5-Max has outperformed GPT-4o and DeepSeek-V3 in multiple tests.

Qwen2.5-Max: Exploring the Intelligence of Large-scale MoE Model | Qwen
https://qwenlm.github.io/blog/qwen2.5-max/

Qwen2.5-Max is an MoE model created by applying supervised fine tuning (SFT) and reinforcement learning with human feedback (RLHF) to a base model pre-trained with more than 20 trillion tokens of learning data, and the number of parameters indicating the scale of the model has reached 100 billion.

Below is an example showing the high performance of Qwen2.5-Max. First, enter a request to Qwen2.5-Max to 'make a minesweeper web application' along with detailed conditions.

Then the code that actually allows you to play Minesweeper was output.

Below is a table summarizing the various benchmark results for 'Qwen2.5-Max', 'Qwen2.5-72B', 'DeepSeek-V3', and 'Llama 3.1-405B'. Qwen2.5-Max recorded the highest scores in all tests.

The graph below summarizes the benchmark results for 'Qwen2.5-Max (red)', 'DeepSeek-V3 (blue)', 'Llama 3.1-405B-Instruct (gray)', 'GPT-4o 0806 (dark beige)', and 'Claude 3.5 Sonnet 1022 (light beige)'. Qwen2.5-Max outperforms GPT-4o and DeepSeek-V3 in multiple tests.

Qwen2.5-Max is available via Alibaba Cloud API . Qwen2.5-Max is also available in Qwen Chat , a chat AI provided by Qwen.