What's this about?
The Financial Times reported that Anthropic's most capable model is not winning users at the rate its benchmark scores would suggest, because a wave of much cheaper competitors is good enough for most of what people actually do. This is the week's dominant business story and it framed almost every other pricing item below.
What exactly is a "frontier model"?
A frontier model is whichever model sits at the current edge of capability — the most expensive and most capable tier a lab sells. The important thing for this story is that frontier models cost far more per unit of work than the tier below them, so a lab betting on the frontier is betting that enough customers have problems hard enough to need it.
What's the community saying?
HN 808 points, 698 comments
- The most-replied comment argues Anthropic treated pricing like a research experiment — searching for the highest number the market would bear — and that this reads as confusing and unsettling to consumers in a way it doesn't to enterprises.
- A large sub-thread is about writing style rather than price: one commenter says they can no longer tolerate the "punchy LinkedIn marketing-team voice" and cannot prompt it away. This thread was busy enough to spawn its own separate HN story (see Claudette below).
- A visible counter-current from heavy users: one on the $200 plan describes an 18-hour autonomous project rebuild and says nothing else comes close, arguing this is a professional tool, not a consumer one.
- A structural read: the frontier model was "too good and given away" on the cheap tier, so the economics now force a push toward paid tokens.
- A non-user's objection on different grounds entirely — unwillingness to hand an organisation's data to any of these companies, citing the book-scanning story below.