Story

Opus 4.5 follows instructions, understands all my tool calls, it understands context, it has a very recent cutoff date... ummm...

I can now one-shot, or two-shot, slightly significant features. I still review all commits line by line, but I find far fewer issues. Am I nuts, or is this like a Sonnet 3.5 level step change?

Of course, anything truly significant requires creating docs, manually reviewing and editing them, then finally implementing, which still has many issues, but the difference in the last few months is blowing my mind and scaring me.

Also, once structured outputs come out of Beta and are available on Azure, I will replace almost every single LLM API call in my own apps with Opus 4.5. Gemini for search grounding, Opus 4.5 for everything else?

edit: May I ask anyone, is Opus 4.5 structered output avail via Bedrock?