Blog post #35

---

Until today I’ve been running the cheapest plan at every major lab — ChatGPT, Gemini, and Claude all on their lowest tier. That’s how you keep exploring without the bill dictating the experiment. But from today I’m on Claude’s max plan, and I wanted to write down why.

---

The decision

Being on the bottom tier everywhere was the right call for a while. I needed to actually use all three to have an opinion, and the cheap seats were enough to form one. Now I have an opinion.

The opinion is: for the next stretch, Claude Code is the strongest tool around. Not the strongest model in every benchmark — that lead trades hands every few weeks and I assume the others will catch up. And probably not the tool that dominates forever; with the source code out in the open, the moat is thinner than it looks.

But right now, today, it leads. And this is where I actually work. So I’m paying for the tier that matches the workload instead of squeezing the lowest one.

What changes in practice

The immediate thing is just — I stop watching the token meter. That low-grade anxiety of “should I really run this with Opus, or switch to something cheaper?” goes away, and the work gets faster because I stop second-guessing every call.

The other labs are still shipping at a pace that keeps me interested. I’ll keep experimenting across all of them — that’s half the point of this blog. But the daily driver deserves the plan that doesn’t make me flinch.

Decisions made

• Claude → max plan. Daily driver tier matches daily driver usage.
• ChatGPT and Gemini stay on the entry tier for now. Enough to keep testing new releases without doubling the stack cost.
• Revisit in a few months. If another lab ships something that genuinely changes where I work, the plan follows the tool — not the other way around.

---

— Stefan