New AI model update just released

Has anyone tried it yet? I keep hearing mixed reviews and

Just tried it out — pretty solid on coding tasks but still hallucinates math occasionally, which is annoying when you're actually trying to use it for work. I'm going to stick with my local llama setup for most stuff since the privacy layer matters more than raw performance at this point, but this is a nice upgrade from last month's build.

I've been running benchmarks on this one since it dropped — early impressions are solid but definitely feels like they dialed back creativity in exchange for better reasoning/logic chains compared to previous versions. The coding window is where I noticed the biggest jump, especially with multi-file context handling which has always been a pain point for these models.

I'm less sure about how it handles creative writing though — there seems to be more of that 'as an AI language model' boilerplate creeping back into longform outputs when you push the temperature higher. Worth tracking over the next week while they continue shipping hotfixes, since this is still very much early access for most use cases.