Yesterday (pro plan) I ran one small conversation in which Claude did one set of three web searches, a very small conversation with no web search, and I added a single prompt to an existing long conversation. I was shocked to see after the last prompt that I had somehow hit my limit until 5:00pm. This account is not connected to an IDE or Code, super confusing.

I cancelled my pro plan last month. I was using Claude as my daily driver. In fact had the API plan also and topped it with $20 more. So it was around $40 each month. Starting from December last year it has been like this. When sessions could last a couple of hours with some deep boilerplate and db queries etc. to architecture discussion and tool selection. Slowly the last two months it just gets over. One prompt and few discussions as to why this and not that and it is done.

I'm guessing their newer models are taking way more compute than they can afford to give away. The biggest challenge of AI will eventually be, how to bring down how much compute a powerful model takes. I hope Claude puts more emphasis into making Haiku and Sonnet better, when I use them via JetBrains AI it feels like only Opus is good enough, for whatever odd reason.

When asking it to write a http library which can decode/parse/encode all three versions of it the usage limit of the day gets hit with one sentence. In the pro plan. Even when you hand it a library which does hpack/huffmann.

please tell me if i'm crazy.

i just refuse to use openai/google/anthropic subscriptions, i only use open source models with ZDR tokens.

- i like privacy in my work, and i share when i wish. somehow we accepted that our prompts and work may be read and moderated by employees. would you accept people moderating what you write in excel, google docs, apple pages?

- i want a consistent tool, not something that is quantised one day, slow one day, a different harness one day, stops randomly.

- unless i am missing something, the closed source models are too slow for me to watch what they are doing. i feel comfortable with monitoring something, usually at about 200-300tps on GLM 5. above that it might even be too fast!

Anthropic went about this in a really dishonest way. They had increased demand, fine, but their response was to ban third-party clients (clients they were fine with before), and to semi-quietly reduce limits while keeping the price the same.

Unilaterally changing the deal to give customers less for the same price should not be legal, but companies have slowly boiled the frog in such a way that now we just go "welp, it's corporations, what can you do", and forget that we actually used to have some semblance of justice in the olden days.

I find Claude code to be a token hog. No matter how confidently the papers say context rot is not an issue I find curating context to be highly important to output quality. Manually managing this in the Claude Webui has helped with my use cases more than freely tossing Claude code at it. Likely I am using both "wrong" but the way I use it is easier for me to reason about and minimize context rot.

Over reliance on LLMs is going to become such a disaster in a way no one would have thought possible. Not sure exactly what, who, when, or where.. Just that having your entire product or repo dependent on a single entity is going to lead to some bad times…

What is the best way to get start with open weight models? And are they a good alternative to Claude Code?

I literally ran out of tokens on the antigravity top plan after 4 new questions the other day (opus). Total scam. Not impressed.

I asked it to complete ONE task:

You've hit your limit · resets 2am (America/Los_Angeles)

I waited until the next day to ask it to do it again, and then:

You've hit your limit · resets 1pm (America/Los_Angeles)

At which point I just gave up

I'm finishing my annual paid Pro Gemini plan, so I'm on the free plan for Claude and I asked one (1) single question, which admittedly was about a research plan, using the Sonnet 4.6 Extended thinking model and instantly hit my limit until 2 PM (it was around 8 or 9 AM).

Just a shockingly constrained service tier right now.

The first hit is free.