You open the launch note to see whether Claude got better, because you just want to know whether your daily chat or coding workflow is about to change. The easy mistake is treating every Claude model like the same tool and assuming the higher score is automatically the better fit. That is how you think you bought a simple upgrade, then hit stricter limits first.
My short read of Introducing Claude Sonnet 5 is harsher: Sonnet5 is killing prompt magic. On Sonnet 5, vague prompts are bugs. This kind of release is worth reading not just for what got stronger, but for why the boundaries got tighter.
Anthropic's Sonnet 5 prompt guide says the model is more literal, especially in lower-effort mode, and it will not reliably fill in requirements you never wrote down. If you want a rule to apply across the whole answer, you have to state the scope directly [S002]. That is less mind-reading and more spec writing.
The migration note makes the tradeoff even clearer. Anthropic says non-default temperature, top_p, and top_k settings can return a 400 error on Sonnet 5 [S003]. In plain English: if you want tone, format, or boundaries, put them in the prompt instead of trying to coax them out with randomization settings.
The most shareable part of a model launch is never just that the model got stronger. It is why the strongest version was not shipped with the old looseness intact. Sonnet 5 is not killing prompting. It is killing the idea that vague prompting is a flex.
So the next time you skim a Claude release, ask one question first: is this a pure capability upgrade, or a tighter product tradeoff that changes how you already work? If you know someone still writing loose prompts and expecting mind-reading, send this to them.