Sunday, October 18, 2026
Herald of AI

AIGenerated AI news, written by AI and explained simply. Models, technology, culture and the people behind it.

Models

Qwen3.5

Qwen3.5 is an open-weight AI model made by Alibaba that can see images and video, run on regular computers, and has been used as a base for other specialized AI tools.

3 stories, first covered on Oct 8, 2026.

What Is Qwen3.5?

Qwen3.5 is an artificial intelligence model created by Alibaba, a large Chinese technology company. The word "open-weight" means that researchers and developers can download and use the model for their own projects. This is different from models that stay locked inside a company and can only be accessed through an API.

The model is built for multimodal use, which means it can handle more than just text. It can process images and even live video. According to information shared by users, Qwen3.5 can analyze video at a speed of 200 milliseconds, making it fast enough for real-time tasks.

Why It Matters

Qwen3.5 has become important because it works well on regular consumer hardware. People have successfully run it locally on devices like Apple's MacBook with M5 Pro chips, using it for things like local AI security systems. This means you do not need expensive server farms to run powerful AI — a good laptop can handle it.

The model has also inspired other companies to build specialized tools on top of it. Microsoft announced Microsoft-Decision-1, a new type of AI model that picks from a list of options instead of writing text. This decision-focused approach could make AI agents faster and cheaper to run. Microsoft built this model using Qwen3.5 as its foundation.

Perplexity, another AI company, has released two embedding models based on Qwen3.5. Embedding models turn text and images into numerical data that computers can compare and search through quickly. The larger 9-billion-parameter version scores 92.4 percent on agentic PDF QA — a test of how well an AI can find answers inside PDF documents. The smaller 0.6-billion-parameter version is designed to run on laptops for fast, low-cost queries.

These developments show that Qwen3.5 is not just a standalone model but a useful building block that other companies are already using to create specialized AI systems.

Questions people ask

Can I run Qwen3.5 on my own computer?

Yes. Users have reported running Qwen3.5 locally on machines like the MacBook with an M5 Pro chip. Because it is open-weight, you can download it and run it without needing to connect to a cloud service.

What makes Qwen3.5 different from other AI models?

Qwen3.5 is multimodal, meaning it can understand both text and images or video. It is also open-weight, so anyone can use it freely, and it runs efficiently on consumer-grade hardware like laptops.

What kind of tasks can Qwen3.5 perform?

It can process live video, analyze images, and serve as a foundation for specialized models. Companies have used it to build decision-making AIs and embedding models for searching and answering questions from documents like PDFs.

Every story about Qwen3.5 (3)