Stability AI co-founder Emad Mostaque teases that an open-source 'fable-level' AI model will run on base MacBook hardware
Story Overview
Emad Mostaque dropped a single-sentence prediction that an open-source model hitting his personal 'fable-level' performance bar will eventually fit on an unmodified MacBook Mini or Air, yet supplied zero architecture hints, size estimates, or even a loose window for when that might happen.
The performance bar stays undefined
Mostaque contrasted the target with 'competent intelligence' like Sonnet but offered no benchmarks, evaluation methods, or examples, leaving anyone curious about real capability without a yardstick to judge progress.
Hardware claims rest on future efficiency gains
The post stresses possibility over immediacy, so readers tracking on-device inference can note the direction but must wait for actual model releases before testing whether base Apple Silicon can deliver without cloud crutches.
Many users expressed excitement about an open-source Fable-level AI model running on base MacBooks because it could democratize access and drive major change, while others dismissed the claims as unrealistic or responded with insults.
No Digg Deeper questions have been answered for this story yet.
Most Activity
My guess is 18 months fwiw.
My forecasts have been reasonably accurate on capabilities previously.
There will be an open source fable-level model that runs on a base MacBook mini / Air or equivalent.
I don’t think people have realised this.
@EMostaque
There will be an open source fable-level model that runs on a base MacBook mini / Air or equivalent.
I don’t think people have realised this.
This is inevitable. The current response of the US government to this is not encouraging.
There will be an open source fable-level model that runs on a base MacBook mini / Air or equivalent.
I don’t think people have realised this.
@scottstts The point isn’t soon. The point is will.
@EMostaque Open fable level model may be 6-9 months away, but running on base MacBook isn’t any time soon
@EMostaque Yup
It will be gradual and slow but it will happen
@EMostaque Qwen 3.6 is old sonnet level and already does, that would be unthinkable just an year and a half
@Rafa_Schwinger Sonnet was actually competent intelligence
Fable was actually decent intelligence
@ramez Do you race to get it out to market based on that? Once they’re in the wild, you’re too far gone
Not sure they can do that on a quick enough time table though
Okay, Mostaque, but who's going to be able to run it? How much VRAM will you need for that? You aren't going to squeeze Fable like intelligence/capability into a a model with only a few billion parameters.
You might, if you have the resources, spend thousands of dollars on compute, GPUs and so on. But distillation and quantization only go so far.
Without some fundamental changes in how the architecture works, I just don't see it. Not any time soon.
@EMostaque Yo I’m all for this but Fable is estimated to be 6T parameter model, even with the best cutting edge quantisation techniques, I don’t think we’ll ever see a Mac mini capable of such compute (unless tensor hardware is revolutionised again) - not to mention minimum RAM needed…
@MrCollison @EMostaque yeah people are too optimistic about performance, like yeah sure models will get more efficient and stuff, but genuinely impressive ones aren’t likely to be ran on a MacBook or even a traditional pc
@EMostaque Because base MacBooks will have the brand new M9 arm chip or will it run on today’s M5?
Small models are catching up but I worry we’re just physics constrained - or I’m missing some fact about the amount of junk in supermassive LLMs
I want this but minimal skepticism warranted
@EMostaque There probably already is. I am betting by end of summer it releases.
@EMostaque there will be 💯, but at this time fable 5 is the charm of AI existience
highist dopamine releaser in human brain 🧠
@EMostaque Intelligence will be everywhere, as will cat videos.
@EMostaque If http://Z.ai does it with a GLM. Microsoft might Copilot it’s own Deepseek V4
@EMostaque That would be wild!!Whats the timeline on that would you suggest? 2029 say?
@EMostaque @BenjaminDEKR No doubt in a year or two we will likely be able to run versions of today’s models however what 10T+ frontier models will be able to do in one to two years may be unimaginable. Today’s models being able to be run locally will be an unlock for sure.
@EMostaque It will be released in Feb 2027. Bookmark this.