v0.41.0 optimizations + Qwen3.5-4B = gamechanger!

[lrq3000]

I tried Qwen3.5-4B with v0.40 and then with v0.41.0, total game changer! The optimizations are really making a HUGE difference! Before the model was not even able to complete a single task, now it is perfectly usable for real tasks and can do complete multiple steps for a single prompt without any hitch.

Qwen3.5-4B (running in Lemonade-server.ai because it is the fastest server I found) is a perfectly capable model to do agentic browser tasks, and it runs very fine on consumer-grade laptops (16GB RAM, RTX 3060 Mobile GPU). Very impressive!

I am going to integrate BrowserOS in my daily toolkit, I often need to scrape data and even though it is much slower than a dedicated script I can make in frontier models, make a scraping script is more cumbersome, so since here there are no token credits being spent with a local model, I am fine with waiting for the task to complete more slowly in the background, because in the end it costs me less time overall (I just have to get to the page I want to scrape and ask the model to do it, this can be much simpler than having to make a scraping script with credentials management).

Also I would suggest to mention than contrary to other similar projects, BrowserOS is the only one that provides full access to the webpage content and DOM to the AI model. Most other similar projects only provide access to what is displayed in the current viewport, which is extremely inefficient.

[shadowfax92]

Thanks so much for the kind words! Really glad R41 is working well for you. Let us know if you spot any issues — happy to add more tools.

We also expose all the tools as MCP so you can use with other coding agents too! - https://docs.browseros.com/features/use-with-claude-code