Post

Sam Rose
@samwho.dev

With Fable being so expensive, and API pricing only, my prediction is that companies will restrict access to only prompts that have been signed off by another engineer/manager. I wonder if PR will come to mean “prompt review” as well as “pull request” 😅

0 reposts28 likes

Andy Bell
@bell.bz

Sorta takes the whole point of the technology away if it doesn’t “save time” tho right? Fable are just an early domino falling as the mass subsidisation comes to an end across the board IMO

Sam Rose
@samwho.dev

I have found that LLMs save me quite a lot of time, personally. I do agree that prices are going up, though. And by a lot.

Jim Jonah
@jimjonah.dev

It feels like it saves a lot of time. But I’ve had it miss, lie, and cheat enough times that my time seems to really be shifted to reviewing and trying to understand code I didn’t write as a defense against that. I listened to a podcast by the author if open code. They sell inference time.

Sam Rose
@samwho.dev

Completely agree. I think I’m probably going to write something on this topic quite soon. I’m coming to the end of a project where I’ve generated tens of thousands of lines of code with AI and painstakingly reviewed all of them. It was an awful slog but necessary. I’m sure few others are doing it.

Jim Jonah
@jimjonah.dev

He claimed they are getting 60-80% margin on that, if I remember correctly, when using open source models. I suspect it’s really recouping the training costs as quick as possible, plus the lack of a quantity of competitors at that level of quality, that’s driving the price increases.

Luke
@lukeocodes.dev

I think the pricing reflects the true cost, and it’s likely driven up by huge amounts of internal reasoning/sub-agent tokens. I predicted Claude 5 would be new architecture, and I can’t help but think 4.7’s shockingly poor performance and 4.8 was the proof of concept for that as core architecture

Aran Wilkinson
@aran.dev

I just can't see how this keeps going cost increase on each new model. At the rate were going these models are just going to be too expensive and no one will be allowed to use them in businesses I'm seeing some engineers spend double the amount of money with less output

Sam Rose
@samwho.dev

LLMs become listed as a perk in job descriptions, individuals get priced out and flock to companies willing to pay for them. Providers offload pesky price conscious B2C customers that they never liked anyway. This frees up inference compute to be used training bigger and better models.