You open Claude at 9:15 a.m. with a client deadline and a long prompt queued. Mid-answer, the session dies with a terse throttle notice. The work you expected to finish this morning is suddenly on someone else’s clock.
I’ve been watching these platform moves for years, and you should treat this as your wake-up call. Anthropic just announced that weekday peak hours will shave through session time faster than before, and OpenAI answered by removing caps on Codex access the same day. I’ll walk you through what changed, who gains, and how this will shape the way you actually use LLMs.
At 5am PT Anthropic changed session behavior
Thariq Shihipar from Anthropic posted that the company will throttle session time during peak weekday windows: 5am–11am PT / 1pm–7pm GMT. Your weekly allowance hasn’t been cut, but the way it’s spread across the week has shifted. That means anyone using Claude between roughly 8am–2pm ET will find sessions end sooner—even on the $100 (€92) per month Max tier.
This isn’t a subtle tweak. Anthropic turned its servers into a tollbooth: the weekly quota remains, but the entry lane is narrower during rush hour. The company says only about 7% of users will hit limits they wouldn’t have before, but that figure masks who gets hurt—people running token-heavy jobs or those who work standard business hours.
How will Anthropic’s session limits affect my workflow?
If you’re running background token-intensive tasks, you’ll see interruptions. Shihipar suggested shifting heavy jobs to off-peak windows, a practical but inconvenient workaround. If your team treats an LLM as a live teammate, this change forces new scheduling habits and introduces delay friction into projects that used to be linear.
At noon, OpenAI announced cap removals for Codex
On the same day, Thibault Sottiaux, engineering lead for Codex, said OpenAI reset usage limits across all plans so people could experiment with recently launched plugins. The message was simple: go build without an artificial ceiling for now.
OpenAI became a carnival barker, offering free rides to draw a crowd. That strategy is obvious: lift caps, lure users back, then decide when and how to monetize. If you’ve migrated between Claude, ChatGPT, Codex, or third-party plugins, you’ll recognize the pattern—temporary generosity to win attention, then a new set of rules.
Can OpenAI’s cap removal win back former users?
Yes, but not guaranteed. Removing caps on Codex makes experimentation frictionless, which attracts builders and teams that value bursty workflows. Companies that rely on predictable, scheduled access will consider cost, stability, and SLAs before returning. The real test is whether OpenAI keeps the generosity long enough to convert exploratory users into paid, consistent customers.
At 8am, teams had to choose between speed and predictability
For product leads and freelancers, this is a new operational constraint. Anthropic’s time-of-day throttling rewards off-hour usage and punishes synchronous collaboration during business hours. OpenAI’s temporary cap removal rewards opportunistic switching and quick prototypes.
Both moves expose a market truth: providers are experimenting with pricing and access mechanics in public. Claude, Codex, ChatGPT, plugins—these are product levers companies use to tune demand and discover monetization points. You will be judged by how fast you adapt: changing schedules, queuing tasks overnight, or migrating models when access is friendlier.
At the keyboard you have three blunt options
Option one: reschedule heavy work to off-peak hours and live with the inconvenience. Option two: diversify across vendors—use Claude when you can, switch to Codex or ChatGPT during peak windows. Option three: lobby for enterprise SLAs and predictable quotas if your work can’t tolerate surprise cutoffs. Each path trades cost, complexity, and reliability.
Practical moves you can make today: batch token-intensive jobs, add retry logic and checkpointing to prompts, and monitor session usage closely. If your team uses plugins or Codex, take advantage of cap-free windows to prototype aggressively—but log everything so the moment caps return you’re not blind-sided.
At boardrooms and Slack channels people will ask what’s next
Expect this to be iterative. Platform operators are balancing demand, compute cost, and market share. Anthropic is smoothing peak load; OpenAI is seeding demand. Both experiments will inform long-term pricing strategies across the industry.
You’re on their schedule now. Will you accept the new timetable, fight for contracted guarantees, or hop between platforms and rebuild workflows each time the rules change?