Opera introduces experimental support for 150 local AI models built into the Opera One browser.

Web browser developer Opera announced in a blog post that it will add experimental support for 150 local major language model variants to its Opera One browser. The models come from about 50 families and will be easily accessible to users through a built-in feature, including Lama by Meta, Vicuña, Gemma by Google and Mixtral by Mistral AI. The rollout is in an experimental phase and is only accessible in the One browser developer flow.

Integrated LLMs

The new feature is part of Opera’s AI Feature Drops program. It is the first time that local LLMs have been added to a major browser in a simple and accessible way. These local LLMs allow users to access these AI models directly from their computers, which means lower latency since the data is not sent over the internet. This also means that data cannot be used to train other models.

Opera One supports a number of local models including Meta’s Lama, Vicuña, Google’s Gemma and Mistral AI’s Mixtral. From this, users can select an LLM, which will then be sent to your computer. Each of these models takes up 2-10 GM of local storage space, meaning they are significantly slower compared to server-based LLMs. The models replace Aria, Opera’s native browser AI.

This is still an experimental version, currently only accessible in the One browser developer flow. Anyone wanting to test the new feature will need to update to the latest version of Opera Developer to gain access to local LLMs.