Anthropic shipped Claude Opus 4.8 this week, the new top model in its lineup. If you use the Claude app, you're probably already on it. It's the default Opus now, and the one running inside Cowork, the mode that builds files for you.
The thing it's better at is long jobs. The kind of work where you ask the AI to do something with twelve steps — research a topic across a dozen sources, draft a report, reason through a problem that doesn't fit in one answer — and you need it to hold the thread the whole way through instead of losing track halfway. That's where 4.8 improved.
On the coding side, the numbers are honest about the limits. Opus 4.8 scored 58% on DeepSWE, a benchmark that tests how well an AI handles real programming tasks. That's about 6 points better than the previous Opus, 4.7, and still behind OpenAI's GPT-5.5. Each task costs roughly $12 to run.
I'll be straight with you: if you don't write code, there's nothing for you to do today. You don't install it, you don't switch anything, you don't change how you work. It's already there.
What you'll feel, eventually, is steadier results on the long stuff. If you've used Cowork to build a spreadsheet model and watched it wobble somewhere around tab four, this is the kind of update that makes that wobble less likely. Open it, give it the same job you gave it last month, and see if it holds together better. That's the whole test.