Claude Code Usage Limits in 2026: What $20, $100, and $200 Actually Get You

Claude Code usage limits in 2026 come in three plans: Pro at $20, Max 5x at $100, and Max 20x at $200 per month. The numbers (1x, 5x, 20x) describe how much more you can do in each rolling 5-hour session compared to Pro, but Anthropic no longer publishes the exact token count that the multiplier is applied to. So the label tells you the ratio, not the actual quantity, and Anthropic reserves the right to change the caps whenever it wants.

---

---

The short version: what each plan gets you

A "rolling 5-hour window" means your usage is measured in a block that starts when you send your first message and resets 5 hours later. The timer follows your account, not the clock, so it does not line up with midnight or the start of the week.

Here is what matters before you pick a plan:

• Pro ($20): one combined weekly cap. Fine for light, occasional coding sessions.
• Max 5x ($100): five times the per-session room of Pro, plus two weekly caps (one across all models, one just for Sonnet).
• Max 20x ($200): twenty times the per-session room of Pro, same two-weekly-cap structure.

Anthropic does not print the baseline number. The only place to see your real remaining limits is Settings > Usage inside the Claude Code app.

Why this matters to you: the plan you pay for guarantees a ratio, not a fixed amount of work. If Anthropic lowers the underlying baseline, your "20x" can quietly buy you less than it did last month.

The comparison table

Token ceilings are approximate, based on third-party math at API list prices. Anthropic does not advertise these figures. The $200 plan credit at list rates works out to roughly 13.3M Opus tokens, 22M Sonnet tokens, or 67M Haiku tokens per month. Credits do not roll over.

The 2026 timeline: four changes in four months

Between March and June 2026, Anthropic adjusted Claude Code limits four times. Each change arrived with little advance warning (reported by users across Anthropic's status page and community forums):

1. March 2026: a prompt-caching bug drained limits 10x to 20x faster than expected. "Prompt caching" is a feature that stores repeated context so you do not pay to re-send it; the bug made it bill as if nothing was cached.
2. April 4, 2026: Anthropic blocked third-party OAuth tokens. "OAuth" is the login method that let tools like Cline, Cursor, and Windsurf use your Claude subscription. After this date, those tools need a separate API key.
3. May 6, 2026: Anthropic doubled the 5-hour limits after securing more compute capacity (reported as the Colossus 1 cluster, via SpaceX).
4. May 13, 2026: a 50% weekly-limit bonus was added, but only through July 13, 2026.
5. June 15, 2026: programmatic usage was moved off the subsidized flat rate onto standard API pricing.

That last one is the big one for builders.

The June 15 billing split, explained plainly

"Programmatic usage" means any Claude Code run that is not you typing in an interactive session. That includes claude -p (a one-shot command), the Agent SDK (code that calls Claude on your behalf), and GitHub Actions (automated jobs that run when you push code).

Before June 15, that automated usage drew from the same flat subscription pool. After June 15, it draws from a separate monthly credit equal to your plan price ($20, $100, or $200). Once that credit runs out, you pay standard per-token API list rates.

Third-party analysis put the effective cost jump at 12x to 150x for heavy automated workloads (reported, not confirmed by Anthropic). If you run agent pipelines or CI jobs against Claude Code, the subscription multiplier no longer protects you once the credit is gone.

The lawsuit

On June 15, 2026, the same day the billing split took effect, a class-action suit was filed in California federal court: Karl Kahn v. Anthropic. The complaint argues the advertised Max plan value was never real. The named plaintiff reported burning 15% of a full Max 20x weekly allowance in a single 5-hour session.

Anthropic's terms of service explicitly reserve the right to adjust caps at its discretion. That clause is now at the center of the case. The outcome is not decided, but the filing makes the core risk concrete: the price tag is fixed, the amount of work it buys is not.

Why owning your harness is the safe hedge

A "harness" is the setup that wraps Claude Code: the agents, skills, hooks, and project rules that turn the model into something that ships real apps instead of code snippets. Two ways to power it:

• Subscription tokens (OAuth): the flat monthly price, but exposed to every cap change above. The April block already proved Anthropic can cut off access overnight.
• Your own API key (ANTHROPIC_API_KEY): you pay per token, you see exact costs, and no interactive-session cap change can touch you. The tradeoff is you give up the subsidized flat rate, which the June 15 change already removed for automated use anyway.

This is the practical reason to own the layer around the model rather than rent it. The $29 Code Kit from Build This Now is exactly that: a complete build system for Claude Code plus a production SaaS skeleton (auth, Stripe payments, PostgreSQL with row-level security on every table). It runs through whichever billing path you choose, so a future cap change does not rug-pull your workflow. The Code Kit is a one-time purchase with lifetime updates, not a subscription.

If you want to go deeper on the pieces that make a harness portable, look at Claude Code subagents, your CLAUDE.md project file, MCP servers, and row-level security. Each one lives in your repo, not on Anthropic's servers, which is the whole point.

FAQ

How many tokens do you get with the Claude Code Max plan?

Anthropic does not publish a token count. Max 5x gives five times the per-session usage of Pro inside each 5-hour window, and Max 20x gives twenty times. The baseline that Pro is measured against is not disclosed. You can only track remaining limits inside Settings > Usage in the app.

What happens when you hit Claude Code usage limits?

Claude Code stops generating and shows a rate-limit message. You wait for the 5-hour window to reset (the timer is tied to your account, not the clock), or if you hit a weekly cap, you wait for your weekly reset date. Switching to API key billing skips the interactive-session caps but charges per-token API list rates.

Is the Claude Max plan worth it?

It depends on how you use Claude Code. For interactive coding, Max 20x meaningfully extends how long you can work before hitting a wall. For automated use (claude -p, Agent SDK, GitHub Actions), the June 15, 2026 change moved that usage to API list rates, so the multiplier no longer applies and you pay per token once the monthly credit runs out.

Can I use a Claude Max subscription with Cursor or Cline?

No. Since April 4, 2026, Anthropic blocked OAuth tokens from Claude subscriptions in third-party tools including Cursor, Cline, and Windsurf. You must use an Anthropic API key for those tools, which bills per token, separate from your subscription.