An 80% token price cut barely changes what a document pipeline costs to run. Max Woolf's post lists every cut since June and asks why. GPT-5.6 Luna went down 80% in July, and Opus 5.5 launched at $4 and $20 per million tokens. The Hacker News thread is about IPOs and open-weight models, not about the apps already in production. I build extraction pipelines for 200-page loan files and for tickets in any layout, and the LLM call was never the slow or the risky step. Most of the engineering goes into retrieving the right ten pages out of two hundred, then checking each extracted field against the page it came from. A wrong number in a loan file costs far more than the tokens that produced it, and that check costs the same at any token price. What the cuts do change is the budget for a second opinion. Sending the hard pages through a bigger model, or through the same model twice, used to be a trade-off and is now cheap enough to be the default. If you run extraction in production, what share of the bill is the model call?