What Changed — May 6, 2026
- Claude Code 5-hour rate limits doubled — applies to Pro ($20/mo), Max ($100–$200/mo), and Team plans, effective immediately
- Peak-hours throttle removed — the 5 AM–11 AM PT weekday limit reduction that launched March 26 is permanently gone for Pro and Max
- Opus API rate limits substantially raised — developers building on Opus models through the API get significantly more headroom
The Background: Six Weeks of Developer Frustration
On March 26, 2026, Anthropic engineer Thariq Shihipar announced that Claude's 5-hour session limits would be reduced during peak hours — weekdays from 5 AM to 11 AM Pacific Time (1 PM to 7 PM GMT) — to manage surging infrastructure demand. The change hit exactly when most developers in the US and Europe do their heaviest coding work. Around 7% of users reported hitting session limits they had never hit before, and developer forums filled with complaints about burning through a full Max 20x weekly allowance by Monday evening.
Anthropic acknowledged the severity on March 31, posting that users were hitting Claude Code limits "way faster than expected" and calling it the team's top priority. The situation was compounded by a separate token-inflation bug introduced in Claude Code v2.1.100 that was silently consuming roughly 40% more tokens per request than expected, and by the April 16 release of Opus 4.7 with a new tokenizer that generated up to 35% more tokens for the same input text.
Today's announcement reverses the most damaging policy change and adds capacity on top.
Change 1: Claude Code 5-Hour Rate Limits Doubled — Pro, Max, and Team
Every subscriber on Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), and Team plans now gets twice the Claude Code allowance within each 5-hour session window. Anthropic previously confirmed to TechCrunch that Pro subscribers could expect 40–80 hours of Sonnet 4 usage per week and that Max 20x offered 20x the Pro allocation. Those multipliers now apply to a doubled baseline, meaning Pro users effectively have the equivalent of what a lightly constrained Max user had in February 2026.
The 5-hour window structure and weekly reset schedule remain in place. The change is to how much you can do inside each window, not to when the window resets.
Change 2: Peak-Hours Throttle Permanently Removed for Pro and Max
The March 26 peak-hour reduction — which caused sessions to burn through the 5-hour window faster during US business hours — is gone. Pro and Max subscribers can now work at consistent capacity regardless of whether they are coding at 6 AM Pacific or 10 AM Pacific. This matters most for developers in Europe and the US East Coast who work standard business hours that align with the old throttle window.
The Free plan situation was not addressed in today's announcement. Free tier subscribers should assume peak-hour behaviour may still vary based on infrastructure load.
Change 3: Opus API Rate Limits Substantially Raised
Developers accessing Opus models directly through the API — rather than through a Pro or Max subscription — get a significant increase in per-minute token throughput. Anthropic has not published the exact new figures, but the announcement describes the increase as substantial. The Opus rate limit on the API is a pooled limit that applies across Opus 4.7, Opus 4.6, Opus 4.5, Opus 4.1, and Opus 4 combined, meaning the additional headroom benefits anyone running any Opus-class model.
For teams running agentic workflows on Opus 4.7 — Anthropic's current flagship released April 16 at $5/$25 per million tokens — this directly reduces the 429 errors that have been appearing during sustained agent loops. Prompt caching continues to be the most cost-effective way to extend effective throughput: cached input tokens do not count toward the Tokens Per Minute limit on most current models, meaning a high cache-hit rate can multiply your effective throughput several times over without changing your plan.
What This Means by Plan
| Plan |
Price |
Rate limit change |
Peak throttle |
| Pro |
$20/mo |
5-hour limit doubled |
Removed |
| Max 5x |
$100/mo |
5-hour limit doubled |
Removed |
| Max 20x |
$200/mo |
5-hour limit doubled |
Removed |
| Team |
$100/seat/yr |
5-hour limit doubled |
No change noted |
| API (Opus) |
$5/$25 per MTok |
Substantially raised TPM |
N/A |
The Token-Inflation Bug: Still Unresolved
Today's announcement does not address the Claude Code v2.1.100+ token-inflation issue that has been burning through quotas roughly 40% faster than expected since late April. The current release as of May 4 is v2.1.126, which patched an Opus 4.7 context-percentage display bug but did not ship a public fix for the underlying inflation. If you are running Claude Code and hitting limits faster than these new doubled allowances would predict, check your version with claude --version. The community workaround is to downgrade to v2.1.34 or reinstall via npm rather than the native binary.
The doubled limits announced today partially offset the token-inflation problem in practice, but the root cause remains open.
Context: Why Anthropic Is Moving Now
Anthropic CEO Dario Amodei has publicly described the company as compute-constrained, with data-center buildouts requiring 18–24 months to translate into available capacity. The March throttle was a demand-shaping measure, not a permanent policy — and today's reversal, combined with a capacity increase, suggests that new infrastructure investment is starting to come online ahead of the originally projected late-2026 timeline.
The timing also follows six weeks of sustained developer criticism, including the April 22 incident where Anthropic briefly removed Claude Code from the Pro plan pricing page before walking back the change after immediate community backlash. Restoring confidence among the developer base appears to be as important to this announcement as the capacity gains themselves.
What To Do Right Now
If you are on Pro or Max and were scheduling heavy Claude Code work around off-peak hours to avoid the throttle, you can stop doing that. The window restriction is gone. If you are an API developer running Opus workloads, recheck your rate limit headers — you likely have more headroom than your current retry logic assumes. And if you are hitting limits that still feel too tight given the doubling, verify you are not running a Claude Code version between v2.1.100 and v2.1.126 without the npm workaround, as the token-inflation bug will eat your gains.
Frequently Asked Questions
Does the rate limit doubling apply to Claude chat as well as Claude Code?
The announcement specifically references Claude Code's 5-hour rate limits. Pro and Max plans share a usage pool between Claude chat and Claude Code, so additional headroom benefits both surfaces, but the primary target of this change is Claude Code's agentic coding sessions.
Is the peak-hours change permanent or another temporary promotion?
Today's announcement removes the peak-hours limit reduction rather than temporarily offsetting it, which is different from the March 13–27 off-peak doubling promotion that expired. Based on the language used, this appears to be a permanent policy reversal rather than a time-limited promotion.
What Opus models benefit from the API rate limit increase?
Anthropic's Opus API rate limit is a combined pool covering Opus 4.7, Opus 4.6, Opus 4.5, Opus 4.1, and Opus 4. Traffic across any of these models counts against the same pool, and the raised limit applies to that combined total.
Does this help with the Claude Code token-inflation bug?
Indirectly — doubled limits give you more buffer against token inflation — but the v2.1.100+ bug itself remains unpatched as of May 6. If your usage still seems disproportionately high after today's changes take effect, the npm reinstall or v2.1.34 downgrade workaround is still the most reliable fix.
Do Enterprise plan subscribers benefit from these changes?
Anthropic's Enterprise tier operates under separately negotiated usage agreements. Enterprise customers should check directly with their account representative for specifics on how today's changes apply to their contract terms.