If you're on Claude Pro or Max and Claude Code has felt tighter lately, you're not imagining it. Anthropic ran a temporary 50% boost to Claude Code's weekly usage limits from mid-May through September 13, 2026. When that boost expired, it was replaced by a permanent uplift of only 25% over the original, pre-boost baseline — which Anthropic itself described as a 17% reduction compared to the level people had been using since May. The limit didn't go up on paper; it went down in practice, for anyone who'd gotten used to the boosted allowance.
This post covers what actually changed, why subscription usage limits move like this in the first place, and what your options are if you're the kind of Claude Code user who runs into the wall before the week resets.
What changed, in plain numbers
Using Anthropic's own framing with a simple index — call the original pre-boost weekly allowance 100:
| Period | Weekly allowance (indexed) | What it means |
|---|---|---|
| Before May 2026 | 100 | Original baseline |
| May 13 – Sept 13, 2026 | 150 | Temporary 50% boost |
| From Sept 14, 2026 | 125 | Permanent 25% uplift over baseline |
Go from 150 back down to 125 and that's a drop of one-sixth — roughly the 17% figure Anthropic used when describing the change. The change applied to Pro, Max, Team, and legacy seat-based Enterprise plans; free accounts and consumption-based Enterprise seats weren't part of the boosted program to begin with. Anthropic hasn't published a universal token or prompt count behind these numbers — by its own account, actual usage depends on conversation length, model choice, tool calls, and the effort level you're running at — so there's no single "you get N messages a week" figure to point to. The 100/150/125 index is a convenient way to describe the shape of the change, not an official unit.
The reaction split roughly into two camps: people who felt the floor move under them mid-workflow and said so loudly (including a wave of cancellation posts when the news broke), and people who pointed out that a temporary promotion ending is not, strictly speaking, a cut. Both things are true at once — nothing was technically taken away that was ever promised permanently, and the number you could actually use dropped by about a sixth.
Why subscription limits move at all
A Claude Pro or Max subscription buys you a share of a pool Anthropic manages across everyone on that tier, not a dedicated allocation. That's a reasonable way to price a product that most people use lightly — but it also means the size of your share is a business decision Anthropic can revisit, the same way a gym membership's "unlimited classes" can quietly become "unlimited classes, off-peak only." The May boost was never framed as permanent; it just felt that way after four months of using it.
This is the same dynamic that makes per-token API billing expensive for coding agents, just one layer up the stack. An agent that re-sends file context every turn, loops through plan-edit-review cycles, and runs for hours at a stretch burns through a shared weekly allowance a lot faster than the "occasional chat question" usage pattern subscription tiers are generally priced around. If your Claude Code sessions are long and frequent, you were always closer to the edge of whatever the current limit happened to be than a casual user — which is exactly who felt the September change first.
Three ways to pay for Claude access, and who controls the ceiling
| Claude Pro / Max subscription | Direct per-token API | Flat-rate gateway (APIClaw) | |
|---|---|---|---|
| Price | $20/mo (Pro), $100–$200/mo (Max) | Pay per token used, no fixed fee | $19–$129/mo by daily request tier |
| Who sets the ceiling | Anthropic, can change it (as in Sept 2026) | No ceiling — bill scales with usage instead | You, by the tier you pick — fixed daily request allowance |
| Can it move without your input | Yes — policy changes apply to existing subscribers | N/A | No — daily allowance resets at 00:00 UTC, doesn't get renegotiated |
| Predictability | Fixed price, variable usage ceiling | Variable bill, no usage ceiling | Fixed price, fixed ceiling |
| Model access | Claude only | Whatever provider you hold a key for | Claude, GPT, Gemini, and more via one key |
The honest takeaway from that table: every option trades one kind of certainty for another. A subscription gives you a fixed bill but a usage ceiling that's ultimately Anthropic's to adjust. Direct API billing gives you no usage ceiling at all, but your bill becomes the variable instead — see the cheapest Claude API breakdown for what that actually costs an agentic workload. A flat-rate gateway fixes both sides: the price doesn't change and the daily allowance is yours for the tier you're paying for, not a shared pool subject to a company-wide policy update.
If you're already hitting the weekly wall
A few concrete options, roughly in order of effort:
- Upgrade tiers. Moving from Pro to Max (or Max 5x to Max 20x) buys more weekly headroom at a higher fixed price. It's the smallest change, and it's still a shared pool subject to the same kind of policy adjustment that just happened.
- Switch to direct API billing for Claude Code. This removes the weekly ceiling entirely, at the cost of a per-token bill that scales with how hard your agent is working — the math in the cheapest Claude API post walks through when that trade-off is worth it.
- Point Claude Code at a flat-rate gateway. This keeps a fixed monthly price but swaps a shared, adjustable subscription allowance for a daily request tier you control. The unlimited tokens API comparison breaks down how "unlimited" is implemented differently across gateways, including on request count vs. token count.
Switching Claude Code to a different endpoint is a configuration change, not a rewrite, since it reads the same environment variables either way:
export ANTHROPIC_BASE_URL="https://apiclaw.biz/v1"
export ANTHROPIC_AUTH_TOKEN="sk-your-apiclaw-key"Full walkthrough, including persisting the config across shells: Claude Code API setup. The same "unlimited tokens per request, fixed daily request ceiling" model is explained in more depth on the flat-rate AI API and unlimited Claude API pages.
When the subscription is still the right call
If your Claude Code usage is genuinely light — a focused session or two a week, well under even the reduced allowance — the Pro or Max subscription is still the simplest option and probably the cheapest one. You also get Claude's consumer-facing features (the chat app, Projects, and so on) bundled into the same price, which neither a direct API key nor a gateway replicates. The September change mainly bites people whose usage had grown to fill the boosted allowance over the summer; if you never noticed it was there, you likely won't notice it's gone either.
The bottom line
Anthropic's September 2026 change was a textbook example of why a subscription's usage ceiling is not the same kind of commitment as its price: the price is contractual, the ceiling is operational, and operational things get adjusted. If you've been pushing against that ceiling often enough to notice when it moved, the more durable fix isn't finding the next limit increase — it's deciding whether you want your usage ceiling to be Anthropic's shared-pool policy, your own per-token bill, or a flat-rate seat you control outright.