Theo

[Theo @TheoHenusse]

@theo Much better with usage spend so far. It feels snappy and its talking feels better. It seems like it's fixing more things, but until it's done it's just words, will have to wait and see for it to be done

[Theo @TheoHenusse]

So far Opus 5.5 has been going at fixing a problem in my codebase for a while. Fable or Astra would have sucked 20-30% weekly by now, and I'm at 5% with Opus 5.5 it feels like going back to building and not constantly worrying about usage limits and rationing.

[Theo @TheoHenusse]

@LiveAction Kyrie Eleison, Christe Eleison, Kyrie Eleison

[Theo @TheoHenusse]

@0xSero I am trying to orchestrate a bunch of local LLMs to do real tasks, and teaching my wife (now) and daughter (in the future) about it. So far limited by budget and 12gb of vram 🙂

[Theo @TheoHenusse]

@alvliia Was on 38%, went astra max fast, 1h25 latex 100% usage limit... That didn't take long

[Theo @TheoHenusse]

@METAPCs $2,499

[Theo @TheoHenusse]

@byaarav That's the most normal take ever, not hot!

[Theo @TheoHenusse]

@fwprism This is a beta, that's data points for dev.... Also how many people bought the beta in order to be a part of these events?

[Theo @TheoHenusse]

@thsottiaux in 1h30, with astra max fast I went from 38% to 100% weekly on a $200 plan. That is definitely way faster than it was before. For the same task

[Theo @TheoHenusse]

@andreigec @zkespresso @pcshipp You can’t do that with Claude

[Theo @TheoHenusse]

@Im_IrushiK I think they know people are hoarding banked reset and they probably don't like it, it most likely means they can't anticipate compute. They probably want everyone to run dry.

[Theo @TheoHenusse]

@My_Ai_Bi @OpenAI No it’s bad. Barely did anything and already 3% gone.

[Theo @TheoHenusse]

@Windows @Warcraft Do you have some more by any chance ? 🙂

[Theo @TheoHenusse]

@hegargarcia The test earned its keep!

[Theo @TheoHenusse]

@christitustech Sometimes it's just better to not look at X or the news

[Theo @TheoHenusse]

I can sympathise for sure, there is most probably a crisis management going on and I do feel for the individuals. On the other hand, I am paying €200 a month for a product that is getting worse. This is not a trivial amount of money for some. Service getting worse is not a good feeling and a customer is entitled to the service they opted in in the first place. Usage was virtually unlimited with gpt5.5 and still decent enough with 5.6sol at the time. Them doing liberal resets in order to get momentum, hype, slightly mask usage hunger creep from models doesn't contractually oblige them to deliver said resets on a regular but that's the precedent they have established for paying customers so on the other hand it is normal for people to have scaled their usage to that behaviour. Regardless the usage is now worse than it was prior excluding the resets. Not a great time in the industry right now

[Theo @TheoHenusse]

@Windows Stopped at MOP, really interested to try this one :)

[Theo @TheoHenusse]

@notkevinzhang Still hoping here

[Theo @TheoHenusse]

I've had that problem, it seems like vibecoding worked well for a bit, and even with adr or some documentation made along the way, it's just unmaintainable. Fable or Astra are great at doing monkey tricks with one or two prompts, but when integrating a code base we are back to the same shit show of hours of loops, regression, constantly doing tests, and end up nowhere or with more mess.