OpenAI researchers reject rumored GPT-5.6 specifications, asserting that new model iterations improve token efficiency

I hate the slop tokenmaxxing reasoning

GPT-5.5 and Fable are great because they don't use a lot of tokens

@iruletheworldmo im just really excited to taste these

@jturntdev nah tomorrow.

@scaling01 how is anthropic ahead here though? both mythos/fable and opus 4.7+ increased their token usage, and use more tokens than their predecessors

anthropic increased token usage (likely because it represents stuff better?), i think it could be a good thing OAI does too?

@iruletheworldmo @jturntdev

@PlastiqSoldier they can, the government just wants to fuck them over because of political disagreements

@iruletheworldmo You also said it was coming today did you not

@scaling01 Opus 4.8 might still end up more expensive than GPT 5.6, especially on cost per task metrics. Anthropic is far behind OpenAI when it comes to cost and efficiency. plus no one will use max reasoning, people will continue to use low->high.

@iruletheworldmo funny, several openai researchers have messaged me saying leo is right on the money

@scaling01 Fable also used double tokens per task than 5.5 on artificialanalysis ranking too..

@scaling01 do you actually think anthropic is ahead in pure reasoning?

@iruletheworldmo No fucking models yet. No 5.6 or Fable.... I am actually sick of Openai and Anthropic.

Anybody know what Mark Zuck is doing? Does he have a good model?

@iruletheworldmo strawberry man… i have supported you for a long time… but if gpt 5.6 doesn’t drop today… i’ve lost my faith in you

@scaling01 More folks should just orchestrate many models

I use 6 providers

@iruletheworldmo Bro I don’t see 5.6 today delete ur account

@scaling01 what if it’s way smarter tho

@blue_square28 chill doggy.

@iruletheworldmo the Leak Squad strikes again

@iruletheworldmo

@SethGammon it’s hot outside bro. enjoy the sun.