The AI Cheapskate

It's now been 1 year since I've been coding around 10 hours a day with AI.

The confession

And it's time to say it publicly: I have never paid for a Max or Pro account at $200/month 😇

On the other hand, I do have basic subscriptions (the ones priced at $20 generally) with pretty much everyone (Claude/OpenAI/Gemini/Warp/ManusAI/Perplexity/Cursor/...), to test and take advantage of each one's specific features.

For the must-have frontier models, I stack them. With Claude Code, for example, I went up to 3 simultaneous accounts last summer, when Codex wasn't quite up to the level yet.

The reason

There was that small friction of logging out of an account that had hit its credit limit, logging into another one, and picking back up the plan that was in progress — but nothing insurmountable.

Financially it was well worth it. And the providers were generous, both on pricing and on limits:

• promotional offers (Claude Code at $9 for 3 months, OpenAI which had 30-day trials at 1 euro at the end of 2025)

• usage extensions (Codex which doubled usage if you went through the app, Anthropic's off-peak hours which consume fewer tokens)

• credits (Cursor's 'legacy' plan which represented 500 requests to a frontier model per month, Claude which offered $40 in credits,...)

The end of an era

In short, all the providers were operating at a loss and we all knew it couldn't last. We made the most of it. We reassured ourselves by saying that open source models weren't that far behind in benchmarks and that we had that safety net if the big players pushed the price dial too far.

By the end of the first quarter of 2026, the acquisition phase seemed to be drawing to a close. The major players are gradually tightening usage.

The models are increasingly powerful, so we waste less time than 6 months ago going back and forth to reach the desired result — which is now generated on the first try — but I rarely get more than 30/45 minutes into Claude before hitting the limit and having to wait for the 5-hour window to reset. Not manageable even with multiple accounts. I'm canceling down to just one.

The era of all-you-can-eat buffets at $20 is over. We still ate pretty well while it lasted.

The Codex killing blow

Codex then remained my safe bet, both for code quality and for the long sessions it allows.

With two accounts and OpenAI's weekly usage resets whenever they had downtime, I managed.

Until early April, when the $20 plan's usage was also reduced, to push adoption of the new $100 pro plan.

Reddit lit up with red 'complaint' badges — just like with Claude, it had become impossible to work a full continuous day with basic plans.

Falling in line

I held out 2 days before upgrading to the $100 Pro plan.
Two days looking at what was available in open source, but the quality was lower — and naturally, as a developer, you always want the best code.

Two days weighing whether it was worth stacking additional basic accounts (going up to 3? up to 4?).

What made up my mind was that on the $100 plan, usage limits are doubled until the end of May 2026.

You can think what you want about OpenAI, but their acquisition strategy is pretty effective. In 2 months they'll have gotten me so used to these new features accessible in the $100 pro plan (sub-agents all the time, extra high reasoning, ChatGPT on model 5.4 Pro which is genuinely impressive,...) that I'll be ready to move to $200. They're good...