ChatGPT's 'Too many requests' error means you've exceeded the platform's message limit for the current time window. The exact trigger and cooldown differ by plan — free users hit limits much faster than Plus subscribers, and GPT-4o has stricter per-window caps than GPT-3.5. This guide breaks down exactly what fires the error and the fastest paths through it.
'You've sent too many messages. Please slow down.' banner below the message box
'Too many requests in 1 hour. Try again later.' error appearing after hitting Send
Message input box grays out or becomes unclickable
Countdown timer showing minutes or hours until the next window
GPT-4o model selector grays out and reverts to GPT-3.5 automatically
'You've reached your GPT-4o limit' even though you're on ChatGPT Plus
Free users get a limited number of GPT-4o messages per session window (roughly 10–20 messages every few hours depending on server load — OpenAI doesn't publish exact numbers). Once the cap is hit, ChatGPT automatically falls back to GPT-3.5 or blocks messages entirely. This is the most common cause of the error.
Plus subscribers have a much higher but still finite GPT-4o limit — approximately 80 messages per 3-hour rolling window as of 2026 (varies with server demand). Heavy users doing long back-and-forth sessions, code reviews, or bulk content generation can hit this within a single work session.
ChatGPT has a per-minute request throttle separate from the hourly window limit. Submitting many short messages in quick bursts — for example, testing prompts rapidly or refreshing a failed generation — triggers 'slow down' errors even when your hourly quota isn't exhausted.
Limits are per-account, not per-device. If you have ChatGPT open in multiple tabs or are logged into the same account on phone and laptop, all usage counts toward the same pool. Heavy usage across devices drains the window faster than you'd expect.
Some Custom GPTs make multiple internal ChatGPT calls per visible response (for example, a GPT that plans, drafts, and revises before showing output). Each internal call burns your quota. A single Custom GPT interaction may count as 3–5 messages toward your limit.
When to try: First option — costs nothing, guaranteed to work
ChatGPT's limits run on a rolling window — usually 1–3 hours. The error message often shows exactly how long until reset. Click the countdown timer or look for the 'Try again in X minutes' text below the input box. This resets automatically; you don't need to log out or do anything. Use the wait to outline your next prompts so you hit the ground running.
When to try: When you need to keep working immediately and the task doesn't require GPT-4o
Click the model selector at the top of the ChatGPT interface and switch from GPT-4o to GPT-4o mini or GPT-3.5 Turbo. These models have much larger (or unlimited) message allowances. GPT-3.5 handles summarization, drafting, simple coding, and Q&A well — save GPT-4o for tasks that genuinely need it. This is the correct workflow for heavy users: use GPT-3.5 for 80% of tasks and GPT-4o for the 20% that need it.
When to try: Specifically for the 'You've sent too many messages. Please slow down.' variant
The 'please slow down' error (distinct from the hourly limit) fires when you send messages too quickly. It clears automatically after 30–60 seconds of no activity. Don't click retry rapidly — that resets the burst timer. Wait a full minute, then send your next message normally.
When to try: If you hit free tier limits more than once a week
ChatGPT Plus ($20/month) gives roughly 4–8x more GPT-4o messages per window compared to free tier. Go to chat.openai.com → click your profile icon → 'Upgrade plan'. If you hit the free GPT-4o limit regularly, Plus pays for itself in saved time within a day or two of moderate use. For teams, ChatGPT Team ($25/user/month) adds per-user higher limits and shared billing.
When to try: When you can't wait and need output now
When rate-limited on ChatGPT, shift to Claude (claude.ai — free tier with generous limits) or Gemini (gemini.google.com — free with Google account). Both run on entirely separate infrastructure with independent limits. Professional workflow: have Claude, ChatGPT, and Gemini all bookmarked — when one throttles you, pivot to the next. Most tasks run interchangeably across the three.
When to try: To prevent hitting limits quickly during intensive work sessions
If you're sending 10 short messages that build on each other, rewrite them as a single well-structured prompt. Example: instead of five messages asking for an outline, then a first paragraph, then a revision — write one message: 'Write a 500-word blog intro on X. Structure: hook → problem → solution preview. Tone: conversational. After drafting, revise for clarity.' One message does the work of five and burns only one unit of quota.
Use GPT-3.5 or GPT-4o mini by default and only switch to GPT-4o for complex reasoning, code review, or nuanced writing tasks
Close extra ChatGPT tabs — usage across all tabs counts against the same per-account limit
Send detailed, complete prompts rather than short back-and-forth messages to get more per token of your limit
Schedule your heaviest AI work sessions at off-peak hours (evenings, early mornings) when server limits are more lenient
Keep Claude and Gemini accounts active as backup options so you never lose productivity to a single platform's limits
The 'too many requests' error is a usage limit, not an account fault — OpenAI support cannot increase your individual limit. However, contact support at help.openai.com if: (1) You're on ChatGPT Plus and hitting limits far below the advertised 80 messages/3 hours, (2) You suspect your account has been flagged for unusual activity causing abnormally strict limits, (3) You're on ChatGPT Team or Enterprise and limits feel wrong for your plan tier. For API rate limits (separate from web interface limits), manage them in your OpenAI API dashboard.
Most message-limit errors reset on a rolling 1–3 hour window. The error message itself often shows the exact time — look for 'Try again in X minutes' or a countdown timer near the message box. The burst-throttle error ('please slow down') clears in 30–60 seconds. If you're seeing limits that last longer than 3 hours, that's unusual and may indicate account-level throttling — contact OpenAI support.
OpenAI doesn't publish exact free-tier GPT-4o limits and adjusts them based on server demand. In practice (April 2026): free users typically get 10–20 GPT-4o messages per session window before being throttled. GPT-3.5 and GPT-4o mini have much more generous (effectively unlimited for normal use) limits on free tier. If you need consistent GPT-4o access, ChatGPT Plus ($20/month) is the practical upgrade.
No, but the limits are much higher. Plus users get approximately 80 GPT-4o messages per 3-hour rolling window (this varies with server load — OpenAI sometimes lowers it during peaks). GPT-3.5 and GPT-4o mini have no practical limit on Plus. If you consistently exceed Plus limits, ChatGPT Team ($25/user/month) provides higher per-user quotas, and ChatGPT Enterprise offers configurable limits.
The 'please slow down' message fires on burst rate (messages per minute), not your hourly quota. If you send 5–6 messages in under 60 seconds — even testing short prompts — it triggers the burst throttle. Wait 60 seconds before sending your next message. This is separate from the hourly limit and clears quickly. The solution is simply to pace yourself: send a message, wait for the full response, then send the next.
No. Limits are per-account, not per-conversation. Opening a new chat does not give you a fresh quota. The rolling window is tracked server-side against your account ID regardless of which conversation you're in. The only thing that resets the limit is time passing (the rolling window expiring) or upgrading your plan.
Each model has separate, independent limits. GPT-4o is compute-intensive and has stricter per-window caps. GPT-3.5 Turbo is much cheaper to run and has far higher (or no practical) limits. When you hit the GPT-4o limit, ChatGPT automatically offers GPT-3.5 as a fallback. For the majority of tasks — writing, summarization, Q&A, basic code — GPT-3.5 output is very close in quality to GPT-4o and burns no quota against the GPT-4o limit.
Yes. Every ChatGPT API call made during a Custom GPT interaction counts against your per-account usage. Some Custom GPTs make multiple calls per user turn (for multi-step reasoning or retrieval). This means a single Custom GPT conversation can consume your limit 2–5x faster than a direct ChatGPT chat. If you're hitting limits unusually fast, check which Custom GPTs you're using — they may be less efficient than direct chat.
Yes, these are separate systems. The web interface limit (covered on this page) is a per-account message count tracked by OpenAI's chat platform. The API rate limit (see our guide at /ai-errors-and-fixes/rate-limit-exceeded) is based on tokens-per-minute and requests-per-minute per API key. Hitting your web interface limit does not affect API calls, and vice versa. Developers using the API manage limits separately in the OpenAI API dashboard. Verified April 2026.