Hacker News new | ask | show | jobs
by agsqwe 369 days ago
thinking models produce a lot of internal output tokens making them more expensive than non-reasoning models for similar prompt and visible output lengths