Alibaba Launches AI Model Qwen 2.5 Max, Outperforms DeepSeek and ChatGPT

Sat, 22 Feb 2025 11:29 | Artificial Intelligence |   Editorial INTI


Alibaba Launches AI Model Qwen 2.5 Max, Outperforms DeepSeek and ChatGPT

Jakarta, INTI – Chinese technology company Alibaba has officially launched its latest artificial intelligence (AI) model, Qwen 2.5 Max, precisely on the Lunar New Year. This AI model is an upgrade from Qwen 2.5, which was released in September 2024.

Compared to its predecessor, Qwen 2.5 Max has significant improvements, particularly in the number of tokens used for training. This model has been trained with over 20 trillion tokens, an increase from the 18 trillion tokens in the previous version. With a larger number of tokens, Alibaba claims that Qwen 2.5 Max outperforms other AI models, such as DeepSeek-V3, OpenAI’s GPT-4o, and Meta’s Llama-3.1-405B.

Qwen 2.5 Max's Superiority in AI Benchmarks

According to an announcement posted via Alibaba Cloud’s official WeChat account, Qwen 2.5 Max has demonstrated superiority across several key benchmark platforms, including:

  1. Arena-Hard
    Score: Qwen 2.5 Max scored 89.4 in Arena-Hard.
    Advantage: This demonstrates excellent ability in following instructions and adapting to human preferences, surpassing other models such as DeepSeek V3, which scored 85.5.
  2. LiveBench
    Score: Qwen 2.5 Max scored 62.2 in LiveBench.
    Advantage: This platform evaluates a model’s overall capability across multiple domains. Qwen 2.5 Max showed better generalization abilities, outperforming DeepSeek V3 by 1.7 points, which scored 60.5.
  3. LiveCodeBench
    Score: Qwen 2.5 Max scored 38.7 in LiveCodeBench.
    Advantage: Qwen 2.5 Max demonstrated stronger coding abilities, surpassing DeepSeek V3, which scored 37.6. This indicates that Qwen 2.5 Max has been well-optimized for coding-related tasks.
  4. MMLU-Pro
    Score: Qwen 2.5 Max scored 76.1 in MMLU-Pro.
    Advantage: It showcases excellent reasoning and knowledge capabilities, with a score nearly identical to DeepSeek V3’s 75.9, proving that Qwen 2.5 Max has very strong cognitive and reasoning skills.
  5. GPQA-Diamond
    Score: Qwen 2.5 Max scored 60.1 in GPQA-Diamond.
    Advantage: Qwen 2.5 Max exhibits better factual consistency in answering general knowledge questions, outperforming DeepSeek V3, which scored 59.1.
  6. New Records
    MMLU-Pro: Qwen 2.5 Max set a new record in MMLU-Pro, showcasing its exceptional reasoning and knowledge capabilities.
    LiveCodeBench: It also set a new record in LiveCodeBench, proving its outstanding coding proficiency.

Intense Competition with DeepSeek

The launch of Qwen 2.5 Max coincides with the rising competition in the AI industry, particularly with DeepSeek, an open-source AI model from China that has gained global attention. DeepSeek is known for its innovative development approach, leveraging open-source technology that allows multiple parties to contribute to its advancement.

DeepSeek emerged amid U.S. trade sanctions against China, restricting the shipment of AI chips to the country. To overcome these challenges, AI companies in China, including DeepSeek, have adopted a resource-sharing strategy and a collaborative approach to accelerate their AI model development. Interestingly, despite being developed with a much smaller budget compared to ChatGPT, DeepSeek remains competitive against AI models from major companies like Alibaba and ByteDance.

How to Use Qwen 2.5 Max

Currently, Qwen does not have a standalone application like ChatGPT but can be accessed via chat.qwenlm.ai. Here’s how to use this AI model:

  1. Visit chat.qwenlm.ai
  2. Click Login and sign in using an email or Google account
  3. Select the desired Qwen 2.5 version, such as Qwen 2.5 Plus or Qwen 2.5 Max
  4. Start interacting with the AI by typing questions or requests

Additionally, Qwen 2.5 offers AI-powered image and video generation features:

  1. Generate images: Type a description of the image, select the "Image Generation" option, and press Enter.
  2. Generate videos: Type a video description command, select "Video Generation," and let the AI process it.

Conclusion

The launch of Qwen 2.5 Max marks Alibaba’s strategic move to strengthen its position in the global AI industry. With superior performance across various benchmarks and a larger training dataset, this model is poised to compete with AI models from OpenAI, DeepSeek, and Meta.

However, competition in the AI industry is becoming increasingly fierce, especially with DeepSeek offering more affordable and efficient AI solutions. Moving forward, innovation and efficiency in AI development will be key factors in winning the artificial intelligence race.

To stay updated on the latest technology event, visit : INTI 2025

 

ArtificialIntelligence MachineLearning AIInnovation +2