Near real-time tool decisions: when you ask for something that needs a tool (web search, opening a page, creating a file...), the assistant now decides what to do in about 1 to 3 seconds, instead of up to 20 seconds before.
Faster web search: searches are about 3x faster, and reading a web page is 2 to 3x faster.
Instant repeats: asking for the same search or the same page again within a few minutes is now instant.
Fast from the very first request: the tool system now warms up in the background at launch, so your first tool request is quick too.
Smoother vision on AMD graphics cards
Less desktop freeze while the assistant reads your screen. This was most noticeable on AMD graphics cards, and the same improvement applies to NVIDIA cards too.
About twice as fast to read a screenshot with the built-in models.
Vision now uses a little more video memory (about 0.5 GB). The app takes it into account automatically when it picks the right settings for your hardware.
Updated built-in AI engine
The built-in AI engine has been updated to a newer version.
Better screen understanding: fewer misidentified games and misread names on screenshots.
Faster processing overall, especially on long conversations.