News and Updates

Qwen3.5 has now been added to ELM® web interface. This state-of-the-art AI model features 397 billion parameters and is hosted entirely by the University. This means it is free to use for all University of Edinburgh staff and students, while keeping your data secure.

API access to Qwen3.5 will be available soon, and we will notify ELM® users when access is enabled.

Below, you will find a simple guide to what this new model can do, how it compares with other models, and why local hosting is such a valuable development for your work and studies.

What is Qwen3.5? (And what are "Parameters"?)

If you are new to AI, think of "parameters" as the connections in an artificial brain. With 397 billion parameters, Qwen3.5 has an incredibly vast library of knowledge and connections. This makes it highly versatile, helping you with everything from writing essays and research proposals to debugging complex code and explaining difficult academic concepts.

Unlike older AI models, Qwen3.5 is natively multimodal. This is a fancy way of saying it doesn't just read and write text: it can understand and analyse images within the very same conversation. It also supports 201 different languages and dialects (including less common ones like Hawaiian and Fijian), making it one of the most globally accessible tools available.                   

How does it compare to Llama3.3?

While Llama3.3 and EuroLLM are highly capable models, Qwen3.5 brings several key advantages to ELM®:

• Visual Superpowers: While Llama3.3 and EuroLLM are primarily text-focused, Qwen3.5 can look at a document, photograph, chart, or screenshot and explain it to you.
• Screen and App Guidance: Qwen3.5 features advanced "visual agent" capabilities. This means it can look at a screenshot of an app or a complex interface and guide you step-by-step on how to use it or complete a task.
• Top-Tier Performance: In academic and reasoning benchmarks, Qwen3.5 matches or exceeds some of the most famous commercial AI systems in logical reasoning, math, and multilingual translation.                         

Why use a "Locally Hosted" model on ELM®?

Most commercial AI tools (like ChatGPT or Claude) are hosted by third-party companies. We do have Zero Data Retention agreements with all our commercial suppliers, but there are still some benefits to using the locally hosted models.

With Qwen3.5 on ELM®, the model is hosted locally on secure infrastructure. This offers several key benefits:

1. Save University money: The Qwen3.5 Opensource model runs entirely on University infrastructure and, unlike commercial models within ELM® like ChatGPT, Claude or Gemini, it does not incur token-based charges to the University. The Qwen3.5 model is equal in power and performance to most of these commercial models, and we are heavily encouraging researchers and staff to use this powerful new model as their default, as it will significantly save money for the University.
2. Additional Security: Any Zero Data Retention Agreement with a provider is a legal protection. With a locally hosted model you have physical protection. User data never leaves the secure University network.  
3. Speed & Reliability: Because it runs on dedicated local systems, it delivers fast, reliable performance.
4. Completely Free to You: While accessing a 397-billion-parameter model commercially would usually require a paid subscription, the University of Edinburgh provides this model to you at zero cost.  
5. Environmentally Friendly: Qwen3.5 combines the deep knowledge and reasoning capabilities of a very large model with the speed and cost efficiency typically associated with a much smaller one. Its Mixture-of-Experts architecture strengthens ELM’s environmental credentials: while the model has 397 billion parameters in total, only 27 billion are active at any given time, significantly improving efficiency. Qwen3.5 is hosted on University of Edinburgh infrastructure powered by renewable energy generated in Scotland, and the datacentre requires no water for cooling. We have closed loop cooling systems in both datacentres where ELM® resides. These systems are highly efficient in their use of power as measured as the ratio of power consumed for work compared to total power consumed to supply and cool and other losses. As a result, Qwen3.5, together with all our locally hosted LLMs, offers an environmentally friendly option for ELM® users.
6. Control over continuity of model lifetime: With the commercial models available under ELM® like ChatGPT, Claude or Gemini, the commercial provider controls when they discontinue any model. We cannot guarantee continuity of commercial LLM lifespan as commercial providers can stop their models being accessed at very short notice. However, with the Opensource models in ELM®, like Qwen 3.5, EuroLLM and Llama – the University is in total control, and we can decide how long a model is available following discussions with you as users.  

How to get started

Using Qwen3.5 is incredibly simple:

1. Log into the ELM® platform using your standard University of Edinburgh credentials.
2. In the model selection dropdown, select Qwen3.5 (397B).
3. Start typing, upload an image, or ask it to help you break down a complex research paper!
4. Whether you are translating historical texts, analysing charts for a lab report, or drafting an email, Qwen3.5 on ELM® is here to help.    

Give it a try today and let us know what you think. Please get in touch with us if you have any questions.                 

ELM® interface update on the 13th August: what you need to know

Following positive feedback on the updated ELM® User Interface (UI) we released on February this year, the original ELM® UI will be retired. This change will provide users with a cleaner, more intuitive web-based experience.

From 13 August, the updated interface will become the default experience for all users, replacing the original version. No action is required: your chat histories, file uploads and other settings will remain unchanged.

The new UI is fully accessible, and it is WCAG 2.2 AA compliant.

If you have any questions or would like support, please get in touch with the team.

Best wishes, 

The EDINA Team 

___________________________________________________________________________________________