Claude answers pricing question by actually running the classification on all three tiers
yoavgo · x · 2026-09-06
Yoav Goldberg shared a delightful model behavior: when asked how much it would cost to classify a document collection with Haiku, Sonnet, and Opus, Claude replied that it was "an important question about tradeoffs" and decided to check rather than guess — then actually ran the classification with each model to produce accurate prices. A viral moment of an LLM being comically thorough.
More from Fun
- Conitzer shares another AI Overview fail: no self-awareness on display — conitzer · 2026-09-06
- "Astra + UE + Blender makes everyone a game dev" meets skepticism: tools aren't the hard part — shekitup · 2026-09-06
- IBM's multi-decade Eames partnership produced a 90ft puppet theater explaining data processing — sarahdrinkwater · 2026-09-06
- Dev asks AI assistant Astra for a second opinion, it instantly spots the implementation bug — antoine_chaffin · 2026-09-06
- Running Codex on a living-room TV with just voice and computer use — ___Patrice___ · 2026-09-06
- Claude Fable has been 'thinking' for over three hours on a three-line prompt — arpitingle · 2026-09-06