Claude Code Rate Limits Doubled: What I Found After Testing Yesterday

---

I almost deleted the notification. Another "we're improving your experience" email from Anthropic, subject line something about higher usage limits — I get those every few months and they usually mean a minor tweak to something I don't use.

But I read it. And the Claude Code rate limits thing was actually real.

I'm Jim Liu, a developer in Sydney running nine websites, including this one. Claude Code is the primary tool in my development workflow — I use it daily for everything from refactoring Next.js data layers to automating content pipelines. Over the past year I've developed an uncomfortably precise sense of when the rate limits kick in. I know roughly what a 90-minute heavy-usage session costs in terms of my 5-hour allocation. I'd started scheduling my most compute-intensive work into the first hour of each session, just to stay ahead of the throttle.

That's a weird way to manage a workday. So I tested yesterday's announcement properly instead of just taking Anthropic's word for it.

What Actually Changed on May 6

Anthropic announced the Claude Code rate limits increase alongside a compute deal with SpaceX. They're accessing Colossus 1 — the same data center that handles xAI's Grok training — adding over 300 megawatts of capacity and more than 220,000 NVIDIA GPUs, coming online "within the month."

The user-facing changes are:

• Claude Code rate limits doubled for the 5-hour window across Pro, Max 5, Max 20, Team (seat-based), and Enterprise (seat-based) plans
• Peak-hours restrictions removed for Pro and Max — those plans no longer get reduced limits during high-demand periods
• API rate limits increased significantly for Tier 1, with input tokens per minute up approximately 1500% and output tokens per minute up approximately 900%

The peak-hours removal matters more to my workflow than the raw limit doubling, and I'll explain why.

The Numbers: Before vs After

Plan	5-Hour Limit	Peak-Hour Restrictions	Notes
Claude Pro	Doubled	Removed	Biggest practical change for individual devs on the base paid plan
Max 5	Doubled	Removed	Effectively 5× Pro base → now 10× after doubling
Max 20	Doubled	Removed	Power-user plan; peak-hours removal here has outsized impact
Team (seat-based)	Doubled per seat	Not mentioned in announcement	Each seat gets doubled 5-hour allocation
Enterprise (seat-based)	Doubled per seat	Not mentioned in announcement	Custom contracts may vary — check with your CSM
API Tier 1	~1500% input TPM increase	N/A	~900% output TPM increase; affects non-Claude Code API usage

Anthropic doesn't publish exact token numbers for Claude Code plans — "doubled" is the official description. From my sessions yesterday, that tracks with reality.

What I Found Testing the New Limits Yesterday

Two sessions on May 6, both on Max 20.

Session 1: A refactor across roughly 40 Next.js files — updating a data fetching layer, fixing type errors, adjusting API response shapes throughout. This kind of session is exactly what used to push me into the throttle zone. Previously I'd hit the slowdown somewhere around 80-90 minutes. Yesterday it ran for around 130 minutes before I finished the task and closed it naturally. That's not rigorous testing, but it's consistent with the "doubled" claim.

Session 2: Writing and deploying content — lighter computation but longer duration. I ran this from about 2pm AEST, which overlaps with US East Coast morning traffic. That timing has historically been the worst for peak-hours throttling. It ran for over two hours without issues.

I can't call the peak-hours change confirmed from one day, but the early signal is exactly what Anthropic said it would be.

Peak Hours Gone: Why This Matters More Than the Raw Limit

If you're regularly hitting the 5-hour limit, doubling it is great — you can roughly double your uninterrupted working time. Straightforward.

But peak-hours throttling has a different effect. It doesn't just reduce how much you can do — it changes when you can work. Sydney afternoon (1pm-6pm AEST) is US morning. That's peak Claude traffic globally. I'd learned to front-load heavy Claude Code work to early morning or late evening to avoid those hours. That's real scheduling overhead.

If the peak-hours restriction is genuinely gone for Pro and Max users, I get back a usable mid-afternoon window. That's worth more to me than the raw limit increase, honestly.

The SpaceX capacity is still rolling out — 300+ megawatts "within the month" means the full hardware isn't online yet. Whether the peak-hours removal holds as usage grows back toward the new capacity ceiling is something I'll be watching. But Anthropic has made a structural commitment here, not just a temporary adjustment.

Who Gets the Most Out of This

You'll notice the biggest difference if:

• You regularly hit the 5-hour Claude Code rate limits mid-task (interrupted sessions, forced context compaction from hitting the wall)
• You work during US business hours or APAC late afternoon that overlaps with US morning
• You're on Max 5 or Max 20, where the base limits were already higher and doubling compounds
• You run long agentic tasks — multi-file refactors, test generation across a codebase, documentation runs

You probably won't notice much if:

• You use Claude Code for short, focused coding queries and rarely approach your limits
• You're on API billing with your own rate management
• You're on Team or Enterprise where peak-hours restrictions weren't mentioned as changing

One thing that hasn't changed: context limits. Session rate limits and context window size are separate systems. If you're hitting context compaction in long sessions, that's not affected by this. I've seen some confusion about this on Reddit since the announcement.

What I'm Changing in My Workflow

Three things I'm going to try differently now:

I'm dropping the "front-load heavy work to early morning" habit. That scheduling quirk existed purely because of peak-hours throttling. If it's gone, I should be able to do compute-heavy sessions at 2pm without the slowdown penalty.

I'm going to let some larger refactors run as single sessions instead of breaking them into two-session chunks. I'd been doing that specifically to stay within the 5-hour limits. With doubled limits, a longer continuous session might be feasible without the interruption.

And I'm going to pay attention to whether the peak-hours removal holds over the next few weeks, as Anthropic's new capacity scales up. The SpaceX hardware is coming online over the next month — that's the real test of whether this is a sustainable change or a temporary honeymoon period.

For the broader picture of how I structure Claude Code work day-to-day, the Claude Code skills guide covers what I've built up over the past year. And Claude Code workflow examples has the session patterns in detail.

FAQ

When did the Claude Code rate limits double?
May 6, 2026. The announcement came alongside Anthropic's compute deal with SpaceX.

Does the free plan get doubled limits too?
The announcement specifically covers Pro, Max, Team, and Enterprise seat-based plans. Free plan users weren't included.

Are the exact token numbers published?
Anthropic hasn't published specific token numbers for Claude Code plans. For API usage, they published specific TPM figures (Tier 1: ~1500% input increase, ~900% output increase).

Is the SpaceX capacity already live?
The rate limit increases are live now. The full 300+ megawatts of compute capacity is coming online "within the month" — the hardware rollout is still in progress.

Did pricing change?
No price changes were announced.

What about peak-hours restrictions for Team and Enterprise?
Anthropic's announcement specifically called out removing peak-hours restrictions for Pro and Max accounts. Team and Enterprise weren't mentioned in that context — worth checking Anthropic's documentation or with your account manager if that matters to you.

---

Jim Liu is a developer in Sydney who runs nine websites including OpenAI Tools Hub. He's been using Claude Code daily since late 2024 for coding, content automation, and development workflows.