Founder NotesBehind the Scenes

Claude Code's Free 50% Ends Sunday — What I'm Doing This Weekend

·4 min read

Deadline indicator: Sunday July 13 at 6PM PDT
Deadline indicator: Sunday July 13 at 6PM PDT

Three weeks ago I quietly noticed my weekly Claude Code budget had shifted from "right at the wall by Friday" to "comfortably under." I assumed I'd just gotten better at batching prompts.

I hadn't. Anthropic had given every paid Claude Code plan a temporary 50% top-up on the weekly limit back in May — separate from the 5-hour rolling window — and I hadn't registered it as a real thing until I read that the top-up expires Sunday, July 13 at 6PM PDT.

For a solo founder shipping two products against a five-day ceiling, that's a meaningful reset. So this weekend I'm not coding. I'm doing the things that eat weekly budget fastest, before the top-up disappears.

I let a temporary capacity boost quietly change how much I shipped in a week. That tells me my throughput is plan-bound, not time-bound.

What's actually reverting

Three things changed for Claude Code in May, and only one of them is going away:

  • May 6 — permanent. Anthropic doubled the 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise. Peak-hour throttling for Pro/Max was removed. Opus API rate limits went up. None of this reverts.
  • May 13 — temporary. A 50% increase to the weekly ceiling across the same plans, originally announced to run through July 13 at 6PM PDT. This is the one reverting on Sunday, July 13.
  • May 15 — one-time. Anthropic reset every user's counters to zero so the new limits felt real. Already happened, doesn't matter now.

I'm on Max 5x. My weekly Opus headroom has been sitting around 75 hours under the bump, and goes back to roughly 50 once 6PM PDT Sunday, July 13 hits. That's a third off what I've been quietly planning around.

The Anthropic-side context, in case it matters for your model of the world: the May 6 upgrades came alongside a compute deal with SpaceX — full use of the Colossus 1 data center, 300+ megawatts, 220K+ NVIDIA GPUs. The May 13 weekly top-up was the cheaper-to-ship signal Anthropic sent into a market where OpenAI's Codex was quietly pulling engineers back across the fence. Sonnet 5 (launched June 30) is the longer answer on the model side. The deadline is the cheaper answer's expiration.

A vertical 5-step checklist: big refactor, test-suite sweep, documentation pass, hold routine work, watch for an extension
A vertical 5-step checklist: big refactor, test-suite sweep, documentation pass, hold routine work, watch for an extension

What I'm doing this weekend, in order

1. The big refactor I kept postponing. Multipage migrations, full-folder rewrites, the kind of work that eats an entire 5-hour window before lunch. Those don't get cheaper after Sunday — they get budgeted after Sunday. So they're happening now.

2. The big test-suite sweep. Generating tests for the parts of the codebase I've been quietly hand-waving over. Output volume on test generation is brutal — every assertion is another paragraph of generation. Same logic as #1.

3. The documentation pass. The site copy, the README files, the on-screen strings I've been deferring. Cheap for hallucination risk, expensive for tokens.

4. Hold the routine work. New feature branches, bug fixes, the daily trickle — those I can do on any plan, on any week. Don't burn Sunday budget on them.

5. Watch for an extension, but don't plan on one. Anthropic gave the bump two months, no extension has been announced as of this writing, and the SpaceX capacity is now online — so the infrastructure reason for the temporary lift no longer strictly holds. I'd bet 60/40 against an extension. If one lands, great; I get a second lap of #1–3 next month.

The actual lesson (for me, anyway)

I let a temporary capacity boost quietly change how much I shipped in a week. That tells me my throughput is plan-bound, not time-bound, and I'd quietly assumed the new ceiling was normal. It wasn't. The plan I actually have is the plan I'm going back to Sunday evening.

If you're using Claude Code as your primary build tool — meaning hitting the wall matters to your timeline — schedule the week assuming the rollback is real. If it doesn't revert, you used a day for low-priority work. If it does, you used a day for high-leverage work. Both fine.

The bigger Q3 planner note, separately: Sonnet 5's intro pricing of $2/$10 per million input/output tokens ends September 1, reverting to standard ($3/$15). If you have any production pipelines running Sonnet 5, that's a separate cost line to model before budgeting the quarter.

Limits roll back Sunday. Make Sunday count.