Greg Kamradt Runs 78K Classifications on GPT-5.6 Luna for $60
AI educator processed 158K requests and 143M tokens overnight using batch mode.
Greg Kamradt reported completing a research project that required 78K classifications. He used GPT-5.6 Luna in batch mode overnight, running 158K requests and 143M input tokens for a total cost of $60. Kamradt added that the work was split into 15 batches of 5K calls each, with results returning in roughly 30 minutes per batch on average. Greg Brockman quote-posted the post, commenting "jevon's paradox at work". In a separate reply, swyx asked if someone has made a codex for batch mode yet.
I had a gnarly research project I needed done - 78K different classifications I let GPT-5.6 Luna run over night, batch mode ($.10/M input!) 158K requests, 143M input tokens ended up being $60 You don't have enough tokens working for you
Combined views
Reactions
3 postsGB SW 
swyx@swyx@GregKamradt i wonder if someone has made a “codex for batch mode” yet
- likes: 29
- replies: 6
- bookmarks: 9
- reposts: 0
GK Greg Kamradt@GregKamradt@swyx yeah - for fyi I did 15x batches of 5K calls each and it took ~30min avg to get a batch back w/ luna
- likes: 1
- replies: 1
- bookmarks: 0
- reposts: 0