Victor Taelin, Higher Order Company founder, sparks debate over whether raw scaling guarantees performance against Anthropic's optimized Fable 5
Story Overview
Victor Taelin, founder of Higher Order Company and creator of open-source tools like HVM and Bend, is drawing attention by probing whether sheer model size alone can overtake Anthropic's carefully optimized Fable 5, especially after community estimates floated a 3-trillion-parameter count for the new release.
Benchmarks leave room for interpretation
GLM 5.2 at 744B parameters has posted competitive coding and agent results against Fable 5, yet no audited head-to-head evals exist and Anthropic has released zero official size details, leaving the scaling question wide open.
Builders test the practical edge
Taelin has highlighted speed gains and bug fixes on his own projects with Fable 5, showing that real-world developer workflows may reward optimization choices beyond raw parameter totals.
Users are excited by speculation that GLM 5.2 could scale past Fable 5 because larger models deliver more knowledge and capability, while others dismiss extreme scale as impractical and accuse Sonnet of being intentionally nerfed.
No Digg Deeper questions have been answered for this story yet.
Most Activity
Sorry if annoying but I really want to push that. GLM is not that far from Opus. If they can close the Fable gap, and that doesn't sound that insane anymore, everything changes. I'd go 100% OSS and never look back...
Can we make that happen?
What could I do to help that happen?
So, Sonnet 5 being worse than GLM 5.2 744B implies GLM 5.2 10T would be better than Fable 5? At the end, it all comes down to scale? Or am I missing something?
The reason Anthropic strikes fear into the hearts of OpenAI TS is precisely the suspicion that no, GLM 5.2 10T would not be better than Fable 5, and neither would GPT 5.5 10T scaling laws optimized for *big* models I suspect "Fable" is not full "Mythos" btw, and more like 3T
So, Sonnet 5 being worse than GLM 5.2 744B implies GLM 5.2 10T would be better than Fable 5? At the end, it all comes down to scale? Or am I missing something?
@VictorTaelin it actually looks better in cursorbench:
@VictorTaelin plenty of reasons to think anthropic intentionally undertrained sonnet 5
@hive_echo what fable is even :|
@VictorTaelin Fable 5 was never confirmed to be 5T+. And according to Dario Fable 5 was this good because of both scale but mostly 2 discoveries they made earlier that year.
@VictorTaelin You are missing something
@teortaxesTex makes sense, ty
but I wonder why Sonnet 5 is so underwhelming then
The reason Anthropic strikes fear into the hearts of OpenAI TS is precisely the suspicion that no, GLM 5.2 10T would not be better than Fable 5, and neither would GPT 5.5 10T scaling laws optimized for *big* models I suspect "Fable" is not full "Mythos" btw, and more like 3T
@VictorTaelin I don't think we can extrapolate that simply but a 10T GLM 5.2 model would indeed be insane and probably would sit at the frontier of LLMs, and it'd also cost a fortune for inference
@VictorTaelin You’re assuming their scaling slopes are equal
@VictorTaelin You are not missing anything.
@VictorTaelin xAI was built on "scale is all you need"
@lisa44Yes perhaps they just lacked the datasets such scale demands
I think scale is most definitely involved, but that is hardly the only difference between the models. Surely they have different training techniques. I've just been reading that there are differences in performance when small models are trained from scratch (better) vs downsized from larger ones. (This was in the context of DPO and infinite loops, but I think the general idea still applies)
@VictorTaelin Would be genuinely surprised if Fable 5 was 10T… probably 5T 500B-ish active or like 2T dense. 10T dense would be nuts…like LLM is dead-end level of news 😒
@VictorTaelin it is in the big model smell league of its own :)
@VictorTaelin GLM is close to Opus on some axes but far spikier in my experience. For example, great at web frontend but much worse at mobile apps. It’s a great model especially for OSS but benchmarks aren’t telling the whole story here.
@VictorTaelin How does it all come down to scale? Is Sonnet 5 smaller than GLM 5.2?
@mimighost008 many things even
@VictorTaelin That's exactly the opposite conclusion I've taken from everything happening for the past 2 months in the world of AI.
In particular once I started seeing what GLM 5.2 1-bit can do.
I think there is a lot more we can do outside of increasing model sizes.