Alibaba Cloud announced the release of the Tongyi Qianwen flagship model Qwen2.5-Max upgrade. The Qwen2.5-Max model is the latest exploration result of the MoE model by the Alibaba Cloud Tongyi team, with pre-training data exceeding 20 trillion tokens. The model shows extremely strong comprehensive performance and has achieved high scores on a number of public mainstream model evaluation benchmarks, surpassing the world's leading open source MoE model and the largest open source dense model.
It is reported that developers can currently experience the model for free on the QwenChat (https://chat.qwenlm.ai/) platform, and enterprises and institutions can also directly call the new model API service through the Alibaba Cloud Bailian platform.
Qwen2.5-Max has demonstrated world-leading model performance in mainstream authoritative benchmark tests such as knowledge (MMLU-Pro, which tests university-level knowledge), programming (LiveCodeBench), comprehensive assessment of comprehensive abilities (LiveBench), and human preference alignment (Arena-Hard). The Tongyi team evaluated and tested the performance of the instruction (Instruct) model version and the base (base) model version of Qwen2.5-Max.
The instruction model is a version of the model that everyone can directly interact with and experience. In benchmark tests such as Arena-Hard, LiveBench, LiveCodeBench, GPQA-Diamond and MMLU-Pro, Qwen2.5-Max is comparable to Claude-3.5-Sonnet, and almost completely surpasses GPT-4o, DeepSeek-V3 and Llama-3.1-405B.
The base model reflects the naked performance of the model. Since the base model of closed source models such as GPT-4o and Claude-3.5-Sonnet cannot be accessed, the Tongyi team compared Qwen2.5-Max with the current leading open source MoE model DeepSeekV3, the largest open source dense model Llama-3.1-405B, and Qwen2.5-72B, which is also at the forefront of open source dense models. In all 11 benchmark tests, Qwen2.5-Max surpassed the comparison model.
Currently, Qwen2.5-Max has been launched on the Alibaba Cloud Bailian platform, with the model name qwen-max-2025-01-25`. Enterprises and developers can call the new model API through Alibaba Cloud Bailian. At the same time, you can also use Qwen2.5-Max on the new QwenChat platform to directly talk to the model, or use artifacts, search and other functions.