Opera One brings 150 local AI models directly into the browser

Running a local large language model has historically been a chore. Between wrestling with terminal commands, installing environment managers, or downloading Ollama, most regular web users simply gave up and stuck with cloud chatbots.

Opera wants to change that. The company just announced experimental support for local LLMs directly inside the developer stream of its flagship Opera One browser, marking the first time a major web browser has built native management for on-device AI.

150 models right in the browser

The rollout comes via Opera's new AI Feature Drops Program. Rather than just offering a token open-source option, the browser is adding support for around 150 local model variants spanning roughly 50 model families.

The lineup includes heavy hitters like Meta's Llama, Google's Gemma, Mistral AI's Mixtral, and Vicuna.

To try it, Opera Developer users simply update to the latest build, enable the feature, and pick the model they want to test. The browser then downloads the files directly to the machine.

These models typically take up between 2 and 10 GB of local storage per variant. Once downloaded, the local model completely takes over your prompts instead of Opera's default cloud-based Aria AI, processing everything entirely on your device with zero data sent to external servers.

Why it matters

While Google and Microsoft are busy locking users into their proprietary cloud AI ecosystems, Opera is taking the opposite route by embracing the open-source community.

For privacy-conscious users, researchers, and developers, this makes local AI genuinely accessible. You no longer need to touch a command line to run Llama or Gemma locally, and your private data never leaves your machine. It also provides an easy way to experiment with different open-source architectures side by side.

We think turning the browser into a local model runner is a clever tactical move for an underdog. Instead of burning billions training its own frontier models, Opera is riding the coattails of open source.

That said, the obvious friction point is hardware. Downloading 2 to 10 GB per model is a hefty ask for smaller hard drives, and running modern LLMs locally without decent dedicated GPUs can lead to agonizingly slow inference times. Opera made the setup frictionless, but local computing limits are still very real.