Default Model Selections - I've been trying haiku-4-5.

[emory]

I occasionally hit a $10 day when I was getting started, but my DA had some great ideas and created some workflows that use The Cheap Seats (Sundae's words 😂)

But I decided to try something a couple weeks ago and set the default model in my .env in ~/clod to the newly released Haiku 4.5 model. Which frankly seems to be every bit as capable as claude-sonnet-4, while being cheaper and faster. I have used Opus exactly 0 times yet, haven't felt like I needed to yet but 4.5 sounds hot as hell so I'll give it a shot soon.

Has anyone else swapped in Haiku by default recently? There have been some errors I spot, and it does happen more often than under sonnet-4.5 but they're usually not a show-stopper and since I don't faceroll and dispatch deliverables via /yolo or anything I've got eyes on it. Anything that involves updating contexts, creating new ones, writing skills or doing integrations, and most of the agents/ in PAI_HOME really are explicitly defining sonnet-4-5, and operations like my DA's [[MOBILE AURA]] being generated use sonnet-4-5 as well.

I wish I had an agent monitoring dashboard like the ones I've seen some of you using that have a timeline view and let you easily open detail on actions taken. It seems like Haiku 4.5 is completely able to be the default though, and my API spending has dropped noticeably without obvious failures.

Also I don't specify jumbo context models at all, I don't do software development at the complexity many of you do either and it hasn't been a problem for me. Not yet at least.