ChatGPT

The August 2025 GPT-5 launch is the pivot: it replaced GPT-4o's looser allowances with per-model three-hour caps, dropping Plus to 80 messages per three hours. Backlash worked — within a month, in September 2025, that cap doubled to 160 and the weekly Thinking limit jumped substantially. After that the pressure turned additive. A February 2026 US test put contextual ads on Free and Go, with ad-free framed as buying back daily messages, and days later the legacy GPT-4o, 4.1 and o4-mini models were retired from the picker, a real hit to anyone standardized on a specific older model. Pricing, meanwhile, eased at the edges: the cheap $8 Go tier went global in January 2026, and a $100 Pro mid-tier slotted in below the $200 plan in April. So limits get set aggressively at model launches, ease under user pressure, and new revenue mechanisms — ads, model retirement, tier proliferation — accrete on top; user reports of inconsistent limits that change without notice and tighten at peak match that. ChatGPT's ceilings are negotiable but volatile. Expect a squeeze whenever a new flagship ships, some relief if enough people complain, and no guarantee that a model you rely on stays selectable.