How Qwen 2.5 Just Beat the Top AI Models — HubSpot SVP of Marketing Shares The Industry Impact

By kflanagan@hubspot.com (Kieran Flanagan)
The AI model races are heating up. Right on the heels of DeepSeek-R1’s release, the industry is reeling from yet another powerful AI model hitting the market. I test drove the latest iteration of Alibaba’s Qwen models — Qwen 2.5.
In this post, I’ll break down what Qwen 2.5 is, how you can use it, and how it compares to OpenAI o1 and DeepSeek-RI. I’ll also explore what this means for the AI industry moving forward. Let’s dive in.
What Makes Qwen 2.5 Different
Qwen 2.5 was released as a surprise launch on January 29, 2025. Like its competitors, Qwen 2.5 offers natural language processing, versatile use cases, and integrations with multilingual support. It’s fast and trained on a massive amount of data. It can search the web, write text, and code.
Unlike OpenAI and Claude’s models, Qwen 2.5 is open source, which opens a realm of possibility for companies and developers.
Beyond that, you can go to Qwen’s website and sign up to start using it today for free. Early testing suggests that Qwen 2.5 performs similarly to ChatGPT’s o1 and o3 models, which cost $200 per month. For a company or an individual looking to leverage complex reasoning and build a custom AI model, that’s significant savings.
Qwen 2.5 is also multimodal, meaning it can process and generate content based on both text and image inputs. This approach makes the tool incredibly versatile. With Qwen 2.5, I can:
- Generate images and videos.
- Create structured outputs for forms and invoices.
- Conduct spacial seasoning tasks.
- Convert images into coding languages like HTML, JSON, and more.
How Qwen 2.5 Compares to Other AI Models
Take a look at this performance comparison of Qwen 2.5 versus the other leading models, including ChatGPT-4, Claude 3.5 Sonnet, DeepSeek-V3, and Llama-3.1.
Qwen outperforms all other models on Arena-Hard (complex problem-solving) and LiveBench (competence in real-world AI tasks). Other tests have found that the model performs better at mathematical reasoning and vision-language modeling, where it needs to process both image and text inputs.
Qwen performs on par or better than paid models from comparable U.S. companies on a variety of tasks. Now, let’s dive into the use cases. Here’s what you can actually do with Qwen 2.5.
Four Ways to Use Qwen 2.5
1. Create images, videos, and text-based content.
First off, Qwen 2.5’s image and video creation rivals DALL-E and Soros. Here’s an AI-generated image someone created of a dog drinking a beer. It’s not perfect, but it’s a decent first take.
Then, for videos, check out this example from Shruti Mishra of a lifelike ride with huskies.