Been testing Opus 4.8 for the past few days.

Anthropic cooked on this one. I'm impressed.
Here are my thoughts on the 4.8 update *so far*
• Pretty slow, but smart. Kind of reminds me of when O3 from GPT first released
• Being able to swap reasoning effort directly in-app is a complete game-changer. I'm using this daily, and "low" as a default level actually seems sufficient.
• 4.8 often says it can't do tasks that I know it can. This is probably due to Anthropic trying to solve the hallucination problem. Kind of annoying.
• Ultracode + 4.8 is crazy good... burns through tokens though (as expected)
• Dynamic workflows are my favorite thing shipped in this entire update. They're super easy to deploy by simply typing "workflow" directly in the prompts.
• Responses from 4.8 in Claude Code and the normal desktop seem vastly different. Almost feels like in the desktop app, 4.8 "hedges" its answers more often and is scared of being wrong.
• I've switched all my workflows from 4.7 to 4.8 - there's no reason not to (same price) and 4.8 is objectively smarter.
Overall, 4.8 experience has been enjoyable thus far, but if I'm being honest, I think 5.6 will crush it...