I had a pretty full day of coding planned. I was able to knock out critical things earlier with Sol and Terra, but now even those are getting overwhelmed it seems.
So… now I’m making some coffee and walking around the office catching up. I might go buy a new chainsaw chain and drop the old one off for sharpening (I accidentally put mine into the dirt the other day cutting up a fallen tree).
Rough day for the Cursor team! Unusual to see three model providers have issues simultaneously. I’d be curious to read the root cause analysis if it’s released. My first inclination is to say it’s always DNS", but we shall see.
Chin up, @Colin , @deanrie , @mohitjain !
Went for a walk and a long chat with Claude on my phone to discuss the meaning of life. 
Hey does anyone have quality drop of auto mode answers? I have seen today after this outage…
Looks like we’re back up and running. Back to the Grok 4.6 grind!
EDIT… and I spoke too soon. It was working fine, but down now. Might just be everyone hopping on at once.
Hey, thanks for your patience and for keeping the thread upbeat. Today’s been a rough day for model availability. Things are already recovering, but like @Warren_James said, when everyone comes back at once there may be short dips. It should settle down.
@Ansh_Triapthi during the overload, some requests get routed to fallback models via Auto, so the answers may have felt different than usual. That’s because of today’s situation, not because the models changed. If you still notice a quality difference after things stabilize, please start a separate thread with a couple of Request IDs and we’ll take a closer look.
And now I’m also worried about the chainsaw chain, @Warren_James.
Thanks, @deanrie – makes sense. I was able to use Grok 4.5 without issue and now it appears that 4.6 is working at normal speed as well. Must have been quite the day (and probably still is) for everyone involved!
EDIT: chainsaw update. purchased a new chain, got the rest of the tree bucked. Also, many hours later and everything is working super smooth as is usually the case.
Beware of customers with chainsaws, I say! 
I’ve been using Opus 5 for the last couple of hours with no errors, unlike this morning.
Incidentally, Ars Technica covered the multi-provider outage yesterday: Four major AI models suffer rare overlapping downtime - Ars Technica
One Arsian’s theory is that it was a domino effect; one model provider / AI assistant goes down, and a mass of traffic got directed to alternative providers, overloading them. We shall see.
I simply suffered through it by watching my requests run very very slow and when they died, I’d just resubmit. Even with that experience, it’s faster then me typing code by hand still. 
playing Zelda