Rationing: The Double Cross

On May 6, Anthropic announced it had doubled Claude Code's 5-hour rate limits for Pro, Max, Team, and Enterprise plans. Peak-hour throttling was eliminated. The announcement cited a new compute partnership with SpaceX's Colossus 1 data center: 300+ megawatts, 220,000+ NVIDIA GPUs.¹ Two days later, developers started reporting the opposite of what was promised.

What They Doubled

Anthropic's announcement was specific: the 5-hour rate limit window was doubled for paid plans. The peak-hour penalty (which had previously halved limits during 5 AM to 11 AM Pacific on weekdays) was removed for Pro and Max users.¹

What the announcement did not mention: the weekly cap. Multiple users noted that while the 5-hour bucket got bigger, the total weekly allocation stayed the same.² The result: you can now burn through your weekly quota in half the time.

What Users Are Reporting

"In the last week, it has been unusable. It is now getting confused, suggests architecture changes that make absolutely no sense," wrote one engineer on Hacker News.³

"I can't complete tasks in a single turn anymore. Claude often stops streaming because it hits some internal tool-call/turn limit."³

"I hit 20% of the weekly limit in about 2 hours on the first day of a new week," reported another user, calculating that at that rate, the weekly limit would be exhausted in 10 hours of work.⁴

A Pro subscriber said they had "stopped using Opus entirely because it's so token-hungry that it drains weekly quota too fast to be practical." The same user noted they now use free alternatives like DeepSeek and Z.ai more than Claude.³

The pattern echoes March 2026, when Claude Code Max subscribers reported sessions draining in as little as 19 minutes. A Reddit thread titled "20x max usage gone in 19 minutes" accumulated over 330 comments. Another, "Claude Code Limits Were Silently Reduced and It's MUCH Worse," gathered 360+ comments in six days.⁵⁶

The Quality Question

Rate limits are one issue. Quality is another. An "Ask HN: Is Claude Getting Worse?" thread surfaced reports that Sonnet 4.6, which had been praised for producing long, correct code in single passes, no longer does so. Users describe needing multiple continuations per task, which burns tokens faster, which hits the weekly cap sooner.³

Some users theorize that Anthropic is running lower-quality quantized models during high-load periods without disclosing the change. Others suspect the 1 million token context window, a flagship feature, is itself part of the problem: the context window consumes capacity that the infrastructure cannot consistently deliver.⁴

Anthropic has not responded to these specific claims.

The Rationing Pattern

This is the eleventh installment in our rationing series. The pattern has been consistent since March 2026:

1. Users report degraded limits or quality.
2. Anthropic acknowledges a capacity problem.
3. Anthropic announces an improvement (off-peak doubling, SpaceX deal).
4. Users report the improvement didn't help, or made things worse.
5. Repeat.

The SpaceX deal was supposed to break the cycle. 220,000 GPUs. The limits doubled. Two days later, the same complaints.

Tell Us Your Experience

Are you a Claude Code user? Has your experience improved since May 6, or gotten worse? We want to hear from developers with specific, dated observations.

What we're looking for:

- Your plan tier (Pro, Max, Team, Enterprise)
- Whether your effective usage improved or degraded since May 6
- Specific examples: tasks that used to work in one turn, sessions that drained faster than expected, quality changes you've observed
- Screenshots of rate limit warnings with timestamps are especially useful

Email bustah_oa@sloppish.com, find us on Bluesky, or reach us on Signal at sloppish.910. We'll follow up with what we find.