181–190 of 695 posts
We're generating all of the code for swamp[1] with AI. We review all of that generated code with AI (this is done with the anthropic API.) Every part of our SDLC is pure AI + compute. Many feature requests every day. Bug fixes, etc.
Never hit the quota once. Something weird is definitely going on.
"effortLevel": "high",
"autoUpdatesChannel": "stable",
"minimumVersion": "2.1.34",
"env": {
"DISABLE_AUTOUPDATER": 1,
"CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": 1
}
1. Nuke all other versions within /.local/share/claude/versions/ except 2.1.34. 2. Link ~/.local/bin/claude to claude -> ~/.local/share/claude/versions/2.1.34
This seems to have fixed my running out of quota issues quickly problems. I have periods of intense use (nights, weekends) and no use (day job). Before these changes, I was running out of quota rather quickly. I am on the same 100$ plan.
I am not sure adaptive thinking setting is relevant for this version but in the future that will help once they fix all the quota & cache issues. Seriously thinking about switching to Codex though. Gemini is far behind from what I have tried so far.
Earlier quoted context omitted.
Where is your evidence of this "massive cost"? Inference is massively profitable for both anthropic and openai. Training is not.
The evidence is that quotas exist, as seen here, and are low enough that people are hitting them regularly. When was the last time you hit your quota of Google searches? When was the last time you hit your quota of StackOverflow questions? When was the last time you hit your quota of YouTube videos? Any service will rate limit abuse, but if abuse is indistinguishable from regular use from the provider's perspective,…
There's a lot of angles you take from that as a starting point and I'm not confident that I fully understand it, so I'll leave it to the reader.
If this helps, I rolled back to version 2.1.34. Here is the ~/.claude/settings.json blurb I added: "effortLevel": "high", "autoUpdatesChannel": "stable", "minimumVersion": "2.1.34", "env": { "DISABLE_AUTOUPDATER": 1, "CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": 1 } I also had to: 1. Nuke all other versions within /.local/share/claude/versions/ except 2.1.34. 2. Link ~/.local/bin/claude to claude -> ~/.local/share/claude/…
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000 export MAX_THINKING_TOKENS=31999 export DISABLE_AUTOUPDATER=1 export CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1
Claude has gotten noticeably worse for me too. It goes into long exploration loops for 5+ minutes even when I point it to the exact files to inspect. Then 30 minutes later I hit session limits. Three sessions like that in a day, and suddenly 25% of the weekly limit is gone. I ended up buying the $100 Codex plan. So far it has been much more generous with usage and more accurate than Claude for the kind of work I do.…
When will people realize this is the same as vendor lock-in?
"Maybe if I spend more money on the max plan it will be better" > no it will be the same "Maybe if I change my prompt it will work" > no it will be the same "Maybe if I try it via this API instead of that API it will improve" > no it will be the same.
Claude, ChatGPT, Gemini etc all of these SOTA models are carefully trained, with platforms carefully designed to get you to pay more for "better" output, or try different things instead of using a different product.
It's to keep you in the ecosystem and keep you exploring. There is a reason you can't see the layers upon layers of scaffolding they have. And there's a reason why after 2 weeks post major update, the model is suddenly "bad" and "frustrating". It's the same reason its done with A/B testing, so when you complain, someone else has no issues, when they complain, you have no issues. It muddies the water intentionally.
None of it is because you're doing anything wrong, it's not a skill issue, it's a careful strategy to extract as much engagement and money from customers as possible. It's the same reason they give people who buy new gun skins in call of duty easier matches in matchmaking for the first couple games.
The only mistake you made was paying MORE, hoping it would get better. It won't, that's not what makes them money. Making people angry and making people waste their time, while others have no issues, and making them explore and try different things for longer so they can show to investors how long people use these AI tools is what makes them money.
When competitors have a better product these issues go away When a new model is released these issues don't exist
I was paying a ton of money for claude, once I stopped and cancelled my subscription entirely, suddenly sonnet 4.6 is performing like opus and I don't have prompts using 10% of my quota in one message despite being the same complexity.
Something similar is happening with GitHub Copilot too. It's impossible to know what a "request" is and some change in the last couple of months has seen my request usage go up for the same style of work. Toss in the bizarre and impossible to understand rate limiting that occurs with regular usage and it's pretty obvious that these companies are struggle to scale.
I find copilot to be much more straightforward, and I can track per request against my credits. Here is the explanation of what a request is: https://docs.github.com/en/copilot/concepts/billing/copilot-...
This clearly isn't true for agentic mode though. This document is extremely misleading. VSCode has the `chat.agent.maxRequests` option which lets you define how many requests an agent can use before it asks if you want to continue iterating, and the default is not one. A long running session (say, implementing an openspec proposal) can easily eat through dozens of requests. I have a prompt that I use for security scanning and with a single input/request (`/prompt`) it will use anywhere between 17 and 25 premium requests without any user input.
Please unsubscribe to these services and see how they perform:
"Maybe if I spend more money on the max plan it will be better" > no it will be the same "Maybe if I change my prompt it will work" > no it will be the same "Maybe if I try it via this API instead of that API it will improve" > no it will be the same.
Claude, ChatGPT, Gemini etc all of these SOTA models are carefully trained, with platforms carefully designed to get you to pay more for "better" output, or try different things instead of using a different product.
It's to keep you in the ecosystem and keep you exploring. There is a reason you can't see the layers upon layers of scaffolding they have. And there's a reason why after 2 weeks post major update, the model is suddenly "bad" and "frustrating". It's the same reason its done with A/B testing, so when you complain, someone else has no issues, when they complain, you have no issues. It muddies the water intentionally.
None of it is because you're doing anything wrong, it's not a skill issue, it's a careful strategy to extract as much engagement and money from customers as possible. It's the same reason they give people who buy new gun skins in call of duty easier matches in matchmaking for the first couple games.
Stop paying more, stop buying these pro max plans, hoping it will get better. It won't, that's not what makes them money. Making people angry and making people waste their time, while others have no issues, and making them explore and try different things for longer so they can show to investors how long people use these AI tools is what makes them money.
When competitors have a better product these issues go away When a new model is released these issues don't exist
I was paying a ton of money for claude, once I stopped and cancelled my subscription entirely, suddenly sonnet 4.6 is performing like opus and I don't have prompts using 10% of my quota in one message despite being the same complexity.
I am tired of all the astroturf articles meant to blame the user with “tips” for using fewer tokens. I never had to (still don’t) think of this with Codex, and there has been a massive, obvious decline between Claude 1 month ago and Claude today.