I've made a 252x return on my domain inspired by https://t.co/kckx2QriAA
https://t.co/FMfU5AoPxn cost $1.40. I am now at $354 just from posting in a founders community that I am in.
Select your pixels and buy a permanent place on the wall!
@VancityReynolds@Wrexham_AFC@firefox So were you weeing or doing naughty things behind the tree?
It’s just a question, please don’t report me to the season ticket police.
@AdiKumarSaroj@lauriewhitwell@TraMufc That’s what they want.
West Ham (for example) have lost 80% of their season ticket holders since moving stadiums.
That freed the supply up and let the owners increase the prices for tourists.
This is the model that all owners want and it destroys clubs.
@ha1331@Sargon_of_Akkad@sovereignbrah@grok I was making a comparison - I said ‘like a desktop’ so that someone that has no idea what a multi-gpu setup is could grasp the concept.
Oh no, you need significantly more.
Currently OpenRouter lets users rent out their hardware and run local models.
I use GLM 5.3.
Input $0.93 / Output $3.16 per M.
The model is comparative to Opus 4.8 and the price of that model is Input $5 / Output $25 per M.
Such a crazy difference!
You can see why Claude want to suddenly ‘stop’ AI development. They’re on the verge of being taken over by open models.
Funnily enough, when GLM 5.2 was announced Claude went to congress to try and ban local LLM models.
The writing’s on the wall, local models will be the standard.
It will just take the reduction of hardware costs to get there.
IMO it is more likely that everyone will run a local model in the future as opposed to a closed cloud model (like Claude).
For example, take security…
A company would much rather keep their sensitive data on local hardware rather than sending it through an API to Claude, OpenAI etc.
Also for the cost of £45k, they can have a model with sufficient usage to run all of their automation needs. It’s a large upfront cost but over 5 years it pays for itself in risk mitigation costs alone.
Yes but you can also use a service like OpenRouter and use other people’s LLM models for a fraction of the cost of a Claude subscription.
For example, GLM 5.3 is insanely good. Comparable to Claude Opus 4.8 (imo) and a fraction of the cost per 1m tokens.
To run a proper rig locally that can handle this would be upwards of $45k.
I do run local models on my Mac Studio (64gb Ram, M1 chip). It’s very good for Open Claw and light operational work.
Basically, you can now buy hardware (like a desktop PC) that runs AI models that are installed on your actual device locally.
These models are becoming increasingly good and just as good as the models that they charge $200 a month for.
So other than the electricity cost (£10 a month), you can have your own AI, on your own computer that is (almost) as good as Claude.