What's this about?
Anthropic's new flagship aims for Fable 5.1-class work with lower token use, faster output and stronger long-task safety checks.
How is it cheaper?
List prices fall to $4/M input and $20/M output, while cache reads drop to $0.20/M. Anthropic says lower token consumption makes a typical workload 40% cheaper than Opus 5.
What's the community saying?
HN 245 points, 75 comments; launch X post 81,000-plus likes.
- Early testers praise clearer prose and codebase-scale work.
- Readers warn that effort levels make cost charts easy to misread.
- Anthropic itself says benchmark margins now overstate some real-world differences.