Runway founder Siqi Chen teases new AI model Fable, claiming it makes GPT-5.5 and Opus 4.8 feel like GPT-3.5
Engineer Nick Dobos agreed, calling older models regressive.
Positive users praise Fable for its superior reasoning and reliability that makes older models feel obsolete, while negative users express frustration with Fable withdrawals along with annoyance at government intervention.
Most Activity
@blader Crazy how weird going backwards feels
Just a month ago these models felt amazing, and now I’m having fable withdrawals
after a few days with fable, going back to gpt 5.5 / opus 4.8 really makes those models feel like using gpt 3.5 or something
@i_am_jeganraj i like how it did what i told it to do and didn't fuck things up
@apoorvdarshan @blader You could get ultracode performance with fable on low and fable would still cost less and use less tokens. Bigger models have more raw intelligence per token is the best way to look at it. For easy things you won't feel much difference though, for hard problems night/day.
@blader I guess everyone is ragebaiting with fable comparisons today
@blader Feel exactly the same way. And now I'm burning through 5.5 Pro calls just trying to get close. I only use opus 4.8 to preprocess things for 5.5 and to get an uncorrelated second option
@blader Literally yes
@blader
@blader It’s crazy to think gpt 3.5 (175 billion params) can legit be outperformed by something less than a tenth the size today. I can’t imagine when we get Mythos level performance to something we could run with reasonable hardware.
@blader While Opus 4.8 was my best friend before I find myself raging at it all day - nothing gets done well - running anything in /goal seems crazy without Fable coordinating.
@blader It’s brutal.
Kind of scary to think what will happen once they have 10+ GW of compute
But I do miss Fabel
@blader The worst is not that fable was so much better, but the fact that the government can so quickly intervene and clip our wings. It just drains a lot of the vibe and momentum.
@blader Yeah - I’m taking my first real day off of work in many months. Working on those models feels like a waste of time.
@blader it is hilarious that people still argue about MMLU scores. once a system actually takes actions and handles state, raw chat models feel like a really fast typewriter.
@blader nah fr
@blader especially feeling it with opus vs fable at 200-300k context
@blader Maybe your work was too above your human baseline capacity if this is true
@blader
@blader Fable was other level
@blader yeahhh very true that
@blader Can I ask what it is you do with LLMs? I ask because apparently you never did anything with them, or at least anything meaningful if you were always having to, seemingly, fix their slop as you've implied.