Post #2152486
2026-04-12 10:55 UTC
Replies (2)
-
@jakobrosin@universeodon.com 2026-04-12 11:06
@klittle667 I switched mine from using Claude Opus 4.6 to using GPT 5.4, and it's clearly stupider. I have to tell it a lot more what to do and to try better. When it was using Opus, it just figured stuff out and did things. For example, it took me just two sentences for it to check me into a flight. I just gave it a number and my name, and it came back with which seat I wanted, and then after that it just sent me the boarding pass. I haven't tried this with the GPT, but I am afraid this would just break, so I am quite annoyed because it was truly helpful and helped me to concentrate on work rather than navigating around accessibility barriers.
-
@jakobrosin@universeodon.com 2026-04-12 11:09
@klittle667 By the way, to try to save costs a bit, you could try to set it up with the GPT codex. This would use your subscription pricing, not the API token-based pricing, so technically you should be able to control the costs a bit more while using a little bit better models.