Anyone using ChatGPT, Claude, Cursor, etc.?
I just saw that OpenAI DevDay keynote... I'm not sure what they're smoking. There was a $200 plan for 20x quota, and they took it off, then brought it back 2 weeks later for same price, but only 10x, then announced a $500 tier for 25x
It's been a revelation having these LLMs available, and if you're already a software engineer it can multiply your productivity a hundred fold... but I don't think it's going to be dominated by these companies for long.
Consider this: you can lease the new $13,000 Mac Studio for 36 months at $257/mnth:
Mac Studio with M5 Ultra chip
36-core CPU, 80-core GPU, 32-core Neural Engine
256GB unified memory
4TB SSD
You could have two of these in your house/office, running decently-sized models at a respectable speed, fully private, for the price of one ChatGPT Pro 500 subscription, and that's in 2026 with peak chip shortages.
A couple more years and I can see people just buying a decent desktop machine to run the models of today.
I just saw that OpenAI DevDay keynote... I'm not sure what they're smoking. There was a $200 plan for 20x quota, and they took it off, then brought it back 2 weeks later for same price, but only 10x, then announced a $500 tier for 25x
It's been a revelation having these LLMs available, and if you're already a software engineer it can multiply your productivity a hundred fold... but I don't think it's going to be dominated by these companies for long.
Consider this: you can lease the new $13,000 Mac Studio for 36 months at $257/mnth:
Mac Studio with M5 Ultra chip
36-core CPU, 80-core GPU, 32-core Neural Engine
256GB unified memory
4TB SSD
You could have two of these in your house/office, running decently-sized models at a respectable speed, fully private, for the price of one ChatGPT Pro 500 subscription, and that's in 2026 with peak chip shortages.
A couple more years and I can see people just buying a decent desktop machine to run the models of today.
Invisible War FTW
Edited 3 time(s). Last edited by Cyberkilla @
Hmm, to be fair, so far that new GPT-6.1 Sol is pretty good. I've only used 2% of my weekly quota in a few hours of messing with a large C++ codebase. I was wiping out the entire week in less than a day with Astra Medium 
They had better not nerf it in a week when the hype dies down, because this is exactly the amount of value I expect for my money
They had better not nerf it in a week when the hype dies down, because this is exactly the amount of value I expect for my money
Invisible War FTW
I expect models will become more efficient, including routing to the right model. The result would be lower cost processing for the vast, vast majority of users' use cases.
I think the silver bullet for these companies will be CAPEX and compute spend to have frontier models to continue to be the leader. The irony is that the farther they go along that never-ending race, the smaller % of users there will be who actually need those incremental improvements.
I'd think if running models locally became common-place, then these companies would keep the newest shiniest version gated behind their paid subscription.
I think the silver bullet for these companies will be CAPEX and compute spend to have frontier models to continue to be the leader. The irony is that the farther they go along that never-ending race, the smaller % of users there will be who actually need those incremental improvements.
I'd think if running models locally became common-place, then these companies would keep the newest shiniest version gated behind their paid subscription.
