Anthropic just shipped Claude Opus 4.8, and it’s a nice reminder that “better AI” isn’t only about bigger numbers on a benchmark. It’s also about control.
The headline change is an “effort” slider on claude.ai that lets you decide how hard the model should work. That sounds small, but it’s practical: quick answers when you’re triaging a problem, deeper thinking when you’re writing customer-facing copy or planning a feature.
They also introduced a faster mode (2.5× speed) and said it’s now cheaper than before. Put those two together and you get a more flexible setup for teams: you can spend your budget where it matters, instead of paying “premium mode” prices for every single question.
The bigger signal is what Anthropic hinted at next: a Mythos-class model preview being used for cybersecurity, with stronger safeguards required before a broader release.
Concrete action: pick one workflow this week (support replies, code reviews, or research) and define two “effort levels” so your team stops overpaying for simple work.