What they’re doing with billions of tokens every week, beats me. But these users absolutely do exist.
If it was a 50x $500 plan it would be a different story.
6.1 will probably perform even better.
Opus 5.5 is better in that regard than Fable 5.1 except for some bugs with more complexity which it was uniquely suited to.
Do things in a sustainable fashion and treat these models like the swappable commodity that they are by using them via abstraction layers. Bonus point is that you might be surprised how much more you can get for your money by intelligently optimizing specific model usage.
Back then the trips were subsidised by VC money, and also the drivers were contractors, so far lower internal costs for Uber.
“Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent.”
Claude has the same personal-professional split. Free, pro and Max have trained by default since Anthropic's policy change in August 2025. API and team plans didn't. Choosing between one and another is also choosing a training default. The subscriptions offers the opt-out setting, Data Controls for ChatGPT and Account settings for Claude, and it's worth checking before paying.
Be a better friend to yourself. Value your time more.
1. In this absolutely wild era of model progress, with frontier and near frontier models leapfrogging themselves seemingly on a weekly basis it's good to get experience with a variety of providers
2. This isn't exactly hard with OpenRouter!
3. I've had very good results having models from one provider critique plans and code written by models from another provider. Can't do that without subs to at least two providers
I think you can just become slightly careless and off it goes. I just have a £20/mo ChatGPT and I'm getting it to do some HomeLab Kubernetes YAML for me that I have no desire to hand-type. I asked it a question and it zoomed off looking at all sorts of websites, running various git commands, doing anything except asking me for an answer to a question.
It ran through about 60% of my 5-hour allocation in those few minutes, which was an unheard-of rate for me before that, and I can imagine if I had zillions of tokens available I just wouldn't have noticed. As it is, my AGENTS.md now exists and is fairly stern.
If it's $1000 per month to make $0 because you haven't found PMF, less obviously a good deal unless you have a pathway to revenue?
I've been dropping the equivalent of a car payment for three months and it's been worth every single penny; even if I never actually sell a product.
https://github.com/ByteTerrace/Puck
It was originally a small project in scope, just a basic GameBoy emulator, a shader VM, and some Azure integrations. Just a bunch of things I wanted to dabble with in order to learn and have fun. Over the past few months it has grown immensely into something resembling an operating system more than a toy engine. There's even a custom DSL for authoring worlds that abstracts over the state/rendering models; it's completely ridiculous!
Will I make any money off the thing? Probably not, don't care. The things I've learned along the way have been worth way more than the amount of money I burned on a few monthly subscriptions. Learning how to interact with agents and treat them like a team was worth it alone; the product I have now is simply a nice bonus.
What intrigues me more is how are they not done building whatever they were building? LLMs have been jaw-droppingly good for a good 6 months now.
The people spending thousands on LLM bills a month should've been done building significant chunks of Microsoft's or Google's software portfolio by now.
I guess what I mean is how is producing code still the limiting factor in whatever they're doing?
I produce a firehose of personal software and I still cannot max out a $20 plan.
Scope creep? When you give a mouse a cookie and all that...