‘The sudden cost spike exposes the fragility of building businesses on subsidized API access. If your margins depend on cheap tokens, you are entirely at the mercy of the labs.’
I’m just here for the fallout. The businesses who jettisoned their talent in favor of cheap AI garbage. The ones who have seen obvious degradation in work quality and stability, but who have tried to ignore or hide it. They now have to face down CFO’s and executive management who see zero returns on the current spend and are being asked to approve extreme budget increases.
A 230% increase isn’t even a big jump. Try 1000%. That may not even be enough to cover the costs and turn a profit for AI companies.
> The ones who have seen obvious degradation in work quality and stability, but who have tried to ignore or hide it. They now have to face down CFO’s and executive management who see zero returns on the current spend and are being asked to approve extreme budget increases.
Hm, there must be businesses like this, but I presume there are many who experience huge returns on current spend - and even 5x costs would be fine, but currently AI prices (given fixed 'intelligence') are dropping like a stone. I can't imagine that the same level of intelligence is going to be more expensive in 12 months.
It's just normal that there's a huge variation of how businesses and individuals adapt to new circumstances, this has been the case forever.
For some firms not having to deal with employees is less do with cost and more to do with control and a freedom from all the extra overhead that dealing with humans and their complex "needs" brings.
Health care, working conditions, the scourge of a human resources department, complaints and liabilities of every sort imaginable.
I personally would pay double or triple the wage of a human worker to not have all that worry, especially in high regulatory juristictions. If the output is faster and even half as good as well, there is no question at all.
Most humans demand too much and bring little to the table. Real talent is very rare and it mostly wants to work for itself if it can.
Big deal, they are bringing 0731 prices to the same level as 1017. Given their compute constraints (thanks to US sanctions) it makes sense to not support too many legacy models anyway.
Over 12 months I'd expect [0] the $/task of completing a benchmarked task to drop to x0.01 of its original price. If providers increase their prices by 230%, that implies I would expect x0.033 the initial benchmark.
I don't know if this even counts as newsworthy, prices are too volatile. Speaking of, wasn't this model released a few months ago? It must be getting close to being obsolete by now.
It's an outdated model that they're increasing the price of so you switch to the newer one essentially, this whole post is clickbait since it doesn't mention that
V4 Flash 0731? Not even 4.1? And anyways, DeepSeek is an open model, plenty of services provide access to it. Also looking at pricing on OpenRouter it looks like Alibaba is actually one of the more expensive providers.
DeepSeek is different company than Alibaba. I'm not sure why prices of one of the secondary (for DeepSeek models) API providers is newsworthy, but I guess people might not realize it is a different company.
The article failed (deliberately?) to mention that 0731 vision exp is two releases old. It's not nearly as efficient or as capable as the newer 4.1 Flash and 4.1 Flash 1017 models.
This move is most likely the DeepSeek team just making efforts to phase it out.
This is no different from AWS charging higher prices for hosting Kubernetes/MySQL/etc. versions that are end-of-life. Keep your stuff maintained and upgrade, to free up the resources.
‘The sudden cost spike exposes the fragility of building businesses on subsidized API access. If your margins depend on cheap tokens, you are entirely at the mercy of the labs.’
I’m just here for the fallout. The businesses who jettisoned their talent in favor of cheap AI garbage. The ones who have seen obvious degradation in work quality and stability, but who have tried to ignore or hide it. They now have to face down CFO’s and executive management who see zero returns on the current spend and are being asked to approve extreme budget increases.
A 230% increase isn’t even a big jump. Try 1000%. That may not even be enough to cover the costs and turn a profit for AI companies.
They're just phasing out an old model. Pricing on current models hasn't changed.
> The ones who have seen obvious degradation in work quality and stability, but who have tried to ignore or hide it. They now have to face down CFO’s and executive management who see zero returns on the current spend and are being asked to approve extreme budget increases.
Hm, there must be businesses like this, but I presume there are many who experience huge returns on current spend - and even 5x costs would be fine, but currently AI prices (given fixed 'intelligence') are dropping like a stone. I can't imagine that the same level of intelligence is going to be more expensive in 12 months.
It's just normal that there's a huge variation of how businesses and individuals adapt to new circumstances, this has been the case forever.
For some firms not having to deal with employees is less do with cost and more to do with control and a freedom from all the extra overhead that dealing with humans and their complex "needs" brings. Health care, working conditions, the scourge of a human resources department, complaints and liabilities of every sort imaginable. I personally would pay double or triple the wage of a human worker to not have all that worry, especially in high regulatory juristictions. If the output is faster and even half as good as well, there is no question at all.
Most humans demand too much and bring little to the table. Real talent is very rare and it mostly wants to work for itself if it can.
What would you say real talent looks like? Is it just about hard work, or something else?
The only problem is that now you can be banned for being cruel to your AI slaves too, the outrage! /s
The issue is our entire economy is being held up by AI hype.
If there's a grand realization that AI isn't magic, let's just say I'm not optimistic.
If there's a grand realization that it IS magic, it doesn't look so good either. Replaced knowledge workers will not be consuming much either.
Big deal, they are bringing 0731 prices to the same level as 1017. Given their compute constraints (thanks to US sanctions) it makes sense to not support too many legacy models anyway.
https://www.alibabacloud.com/help/en/model-studio/model-pric...
Over 12 months I'd expect [0] the $/task of completing a benchmarked task to drop to x0.01 of its original price. If providers increase their prices by 230%, that implies I would expect x0.033 the initial benchmark.
I don't know if this even counts as newsworthy, prices are too volatile. Speaking of, wasn't this model released a few months ago? It must be getting close to being obsolete by now.
[0] https://artificialanalysis.ai/trends#efficiency
It's an outdated model that they're increasing the price of so you switch to the newer one essentially, this whole post is clickbait since it doesn't mention that
V4 Flash 0731? Not even 4.1? And anyways, DeepSeek is an open model, plenty of services provide access to it. Also looking at pricing on OpenRouter it looks like Alibaba is actually one of the more expensive providers.
I'd like to see the prompt that generated this piece
edit: these are not the prices I see at https://api-docs.deepseek.com/quick_start/pricing/
DeepSeek is different company than Alibaba. I'm not sure why prices of one of the secondary (for DeepSeek models) API providers is newsworthy, but I guess people might not realize it is a different company.
The article failed (deliberately?) to mention that 0731 vision exp is two releases old. It's not nearly as efficient or as capable as the newer 4.1 Flash and 4.1 Flash 1017 models.
This move is most likely the DeepSeek team just making efforts to phase it out.
Yep, 1017 is available at the normal $0.15/$0.06 pricing: https://www.alibabacloud.com/help/en/model-studio/model-pric...
This is no different from AWS charging higher prices for hosting Kubernetes/MySQL/etc. versions that are end-of-life. Keep your stuff maintained and upgrade, to free up the resources.
AI blogspam.
In other news, China is about to ban open weights model exports
https://www.straitstimes.com/business/china-plans-ai-export-...
So reliant on AI
I mean, even with a hike of 230% this is going to be lower than Luna/Haiku right?
Afaik Luna is a bit cheaper.