Mumbai, January 29: China-based Alibaba recently launched its new AI model, the Qwen 2.5 Max AI model, that challenged its rival DeepSeek by scoring better benchmarks compared to its leading model. The artificial intelligence model was launched as a large-scale MoE (Mixture of Experts) model that used machine learning and went through pre and post-training.

With the release of the latest Qwen 2.5 Max AI model, Alibaba shared benchmarks that showed that it had defeated models like DeepSeek V3 and Llama 3.1. The company said that the large-scale MoE LLM was pre-trained on massive data and post-trained with the curated SFT (Supervised fine-tuning) and RHLF (Reinforcement learning from human feedback). Alibaba Qwen2.5 Max model achieved competitive performance against the leading AI models. It outperformed DeepSeek V3 in Arena Hard, LiveBench, LiveCodeBench and GPQA-Diamond benchmarks. Qwen2.5 Max: Alibaba Launches New AI Model, Claims It Surpasses DeepSeek V3, GPT-4 in Multiple Benchmarks.

Qwen 2.5 Max Key Performance Highlights Compared to Other Models

In terms of performance, Qwen2.5 Max scored the highest 89.4 scores in the Arena-Hard benchmark, achieved 76.1 points in MMLU-Pro benchmarks, which are behind Llama-3.1-405B model with 78.0 points and 60.1 points in the GPQA-Diamond benchmarks, which was again below Llama model with 65.0 points. In the LiveCodeBench scores, Qwen2.5 Max scored 38.7 and 0.2 behind Llama-3.1-405B and the highest 62.2 points in the fifth LiveBench test. With all of these scores, the overall performance of the Alibaba AI chatbot was remarkable. 

Qwen2.5-Max Performance Benchmarks (Photo Credits: GitHub)

Qwen2.5 Max Vs DeepSeek V3 Benchmarks 

Alibaba's Qwen2.5 Max outperformed the other models in terms of multiple benchmarks such as MMLU, MMLU-Pro, BBH, C-Eval, HumanEval, CRUX-I, CRUX-O and other benchmark datasets or evaluation methods. It scored highest compared to the Qwen2.5-72B, DeepSeek V3 and Llama-3.1-405B models. 

Qwen2.5-Max Benchmark Datasets (Photo Credits: GitHub)

Qwen 2.5 Max Availability

Qwen 2.5 Max is available on Qwen Chat, allowing the users to directly chat with the artificial intelligence model, search or play with artifacts. It is also available on the popular developer platform GitHub. OpenAI DeepSeek Rivalry: ChatGPT-Maker Probing if Chinese Company Used Its Chatbot Data To Train AI Models Amid US-China Trade War.

Qwen 2.5 Max (Qwen2.5 Max) is another Chinese AI chatbot launched right after the success of the DeepSeek R1 model. However, Alibaba addressed the DeepSeek V3 model and said its new large-scale MoE LLM artificial intelligence model was better. 

(The above story first appeared on LatestLY on Jan 29, 2025 08:49 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).