I use the (now deprecated?) opus plan mode still. Sonnet 4.5 turns in less thought out plans than Opus.
I wish codex cli tool wasn’t so super basic. Not even dedicated plan mode. Very annoying how hard the models are tied to their specific CLI tools.
You say your team uses Sonnet 4.5 more than Opus. However, we pay $200/mo for Opus access (with the reset timer) and now you're forcing people without telling us about the changes you made to Opus usage. You want us to pay for API access, okay. I get it, you need to make money.
But in my case and most others, Sonnet 4.5 is worse than Opus on many tasks and ends up destroying the codebase in many cases. We might not be building AI core companies, but we have complex code bases too and we find Opus works better without making comically bad mistakes.
I'd highly encourage others to try out OpenAI's codex. It works better than Sonnet 4.5, and maybe on par with Opus. At least they don't knee cap usage like Anthropic.
I'm disappointed in how this whole situation has been handled/communicated.
But in my case and most others, Sonnet 4.5 is worse than Opus on many tasks and ends up destroying the codebase in many cases. We might not be building AI core companies, but we have complex code bases too and we find Opus works better without making comically bad mistakes.
I'd highly encourage others to try out OpenAI's codex. It works better than Sonnet 4.5, and maybe on par with Opus. At least they don't knee cap usage like Anthropic.
I'm disappointed in how this whole situation has been handled/communicated.