Breaking News

Claude Code Rate Limits Just Doubled. Here’s What Changed for Your Plan.

May 7, 2026 · 12 min read

Claude Code 5-hour limits doubled across Pro, Max, Team, and Enterprise — May 6, 2026. Weekly limits unchanged.

If you’ve hit Claude Code’s session limit mid-workflow this week — that cap doubled today. Anthropic made the change permanent on May 6, 2026, effective immediately across every paid plan tier.

A common concern among operators who’ve followed this announcement: does doubling the 5-hour limit actually mean more usable time, or is the weekly cap still the real constraint? Here is the exact breakdown of what changed, what didn’t, and whether the math shifts for your specific plan.

What Actually Changed on May 6

Anthropic announced three simultaneous permanent changes at its Code with Claude developer conference in San Francisco. First: the five-hour rate limit for Claude Code is doubled across Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), Team, and seat-based Enterprise plans. Second: peak-hour throttling — the mechanism that caused limits to burn faster during weekdays 5am–11am PT — is permanently removed for Pro and Max accounts. Third: Opus API rate limits were substantially raised, with updated figures now live at platform.claude.com/docs/en/api/rate-limits.

These changes were not conditional on a future infrastructure milestone. They went live on May 6, 2026. The Colossus 1 compute capacity is the backing commitment — not the activation trigger. If you hold an active paid plan, the new limits are already on your account.

One important note on absolute limits: Anthropic confirmed the doubling but did not publish the specific new token or message counts per plan tier. The five-hour window structure itself is unchanged — what changed is the headroom within each window. The relative multipliers (Max 5x is 5x Pro, Max 20x is 20x Pro) remain the same framework, now applied to a doubled Pro baseline.

What Did NOT Change — The Weekly Cap

This is the fine print that matters for operators running sustained daily sessions. The weekly aggregate limits were not part of the May 6 announcement. The doubling applies to how much you can do in a single five-hour window — not to your total weekly ceiling. A developer working six hours a day, five days a week, who was previously hitting weekly limits will hit them at the same point in the week.

Developer forums surfaced this gap within hours of the announcement. One Hacker News commenter put it plainly: the doubling reduces per-session friction but does not add more total monthly capacity for operators who burn through weekly allocations. A Max 20x subscriber who could previously run four full sessions per week before hitting the weekly ceiling can now run those four sessions with less mid-session interruption — but still four sessions, not eight.

The operators who gain the most from May 6 are those whose primary bottleneck was the five-hour window itself — agentic loops stalling mid-task, peak-hour degradation during business hours, sessions burning through limits in 15–30 minutes. Before May 6, Pro users in agentic workflows regularly reported exhausting the five-hour window inside half an hour. That specific friction is meaningfully reduced. If your workflows were already constrained by weekly totals rather than per-session limits, the practical day-to-day change is smaller.

Claude vs ChatGPT at Your Price Point Right Now

The $100/month tier is where the May 6 changes most visibly shift the competitive comparison. OpenAI launched its ChatGPT Pro plan at $100/month on April 9, 2026, explicitly targeting Claude Max 5x. After May 6, Claude Max 5x’s doubled five-hour limits and permanently removed peak-hour throttling make it arguably more predictable for business-hours workflows than ChatGPT’s dynamic throttling system — which applies limits variably based on demand.

At $200/month, the honest assessment has not changed: ChatGPT Pro’s “unlimited” framing still holds a structural edge for the heaviest single-user sustained workloads, particularly uninterrupted Codex sessions. Claude Max 20x is now meaningfully more competitive than it was on May 5, but the weekly cap means Claude is still not “unlimited.” The operator choosing between them at $200 should weigh predictable flat limits (Claude) against an uncapped ceiling with dynamic throttling (ChatGPT).

For a deeper look at how Claude and ChatGPT compare across operator use cases, read our breakdown: Claude vs ChatGPT for Business: The Honest 2026 Breakdown.

The SpaceX Context in One Paragraph

Anthropic’s rate limit improvements are backed by a new compute deal giving Anthropic exclusive access to Colossus 1 — a 785,000-square-foot data center in Memphis, Tennessee, originally built by xAI to train Grok. The facility currently houses 220,000+ NVIDIA GPUs (150,000 H100s, 50,000 H200s, 30,000 GB200s) and 300+ megawatts of power capacity. Anthropic is renting the entire facility under a rental agreement — not purchasing it. SpaceX retains ownership. The capacity is expected fully online by end of May 2026. The rate limit increases on May 6 were not contingent on Colossus 1 coming fully online — they represent a commitment backed by incoming capacity, and they are live now.

What This Means for Your Workflow This Week

The practical impact depends on where your current bottleneck sits. If you run agentic loops — multi-step automations where Claude Code re-reads context at every step, or pipelines running 4–8 concurrent sessions — you were the most likely to hit the five-hour limit before completing a meaningful chunk of work. Those workflows get the most immediate benefit from the doubling. The peak-hour throttle removal means you can run the same workflow at 10am Tuesday as you could at 2am Saturday — without burning through limits faster during business hours.

If you primarily use Claude Code for research, document analysis, or single-session coding work that already completed within the five-hour window, the change is less operationally significant. Your sessions were not being interrupted. The doubled window gives you additional headroom, but it does not change completed workflows that already fit.

For Claude Code at the $20 Pro tier specifically: before May 6, Pro users in agentic workflows reported burning through the five-hour limit in 15–30 minutes. That figure was not unusual — agentic loops with full-context reloads are token-intensive by design, and the Pro baseline was the thinnest. The doubling materially changes the calculus at $20/month for operators who were using Pro as their primary automation platform and hitting limits before lunch.

If you are evaluating whether the $25-per-month difference between Pro and the $45 ChatGPT equivalent makes sense for your workload, this is also relevant context: read our analysis of The $25 Tool That Replaces Your $2,500 Virtual Assistant for the full operator-level comparison.

Disclosure: FutureAIStack may earn a commission on qualifying purchases through affiliate links. Our analysis is independent and not influenced by affiliate relationships.

Frequently Asked Questions