ChatGPT limits feel inconsistent or tighter
Recent tracked changes that can explain it:
What’s going on
ChatGPT's limits feel inconsistent because there is no single global cap — they're separate rolling windows applied per model, with harder throttling layered on when traffic is high. On Plus, the main model runs on a roughly three-hour window (currently around 160 messages) while the Thinking model carries its own weekly ceiling, so which model you're in decides when you hit a wall. That same per-model design gives you a quick read: run out on one model, switch to another, and if you can keep working immediately, you met that model's cap rather than an account-wide block or an outage. The numbers have moved, too — when GPT-5 launched in August 2025 the Plus three-hour cap opened at 80 messages and doubled to 160 a month later after backlash — so a limit that feels unfamiliar may just be a revision. Peak-traffic throttling supplies the daily variability: tighter during busy hours, looser off-peak, which is expected rather than a fault. Two more changes can pass for tightening. Contextual ads now appear on the Free and Go tiers in a US test, where going ad-free means fewer daily messages; and several legacy models (GPT-4o, 4.1, o4-mini) were retired from the picker in February 2026, so a model you used to pick may be gone rather than restricted. So check which model you're on before concluding the limit shrank, keep a second model in reserve for when your primary is spent, and let the rolling window reset instead of retrying into a wall. If one specific model keeps running dry sooner than it used to, that tier has genuinely tightened — but switching models and pacing heavy sessions across the window stays the fastest daily workaround.
ChatGPT has eased its constraints recently.
Low recent constraint volatility — few meaningful restrictions in the last 12 months.
Related problems