ChatGPT · Rate Limits

Chatgpt Rate Limits

ChatGPT consumer/business tiers throttle frontier-model usage via "messages per N hours" caps which OpenAI documents on the pricing/help center pages and may adjust dynamically. The OpenAI Platform API enforces per-organization tier-based limits across requests-per-minute (RPM), tokens-per-minute (TPM), and tokens-per-day (TPD); customers progress through usage tiers (Free, Tier 1-5) as cumulative spend grows. Limits vary per model.

Chatgpt Rate Limits is the machine-readable rate-limit profile for ChatGPT on the APIs.io network, conforming to the API Commons Rate Limits specification.

It captures 6 rate-limit definitions, measuring messages, tokens_per_minute, requests_per_minute, and tokens_per_day.

The profile also includes 4 backoff/retry policies defined and response codes documented for throttled, unauthorized, and serviceUnavailable.

Tagged areas include Rate Limiting, Artificial Intelligence, and LLM.

6 Limits Throttle: 429
Rate LimitingArtificial IntelligenceLLM

Limits

ChatGPT Free (frontier model messages) account
messages
see help center; rolling-window cap on GPT-4o-class messages
ChatGPT Plus / Pro (frontier model messages) account
messages
see help center; higher caps for Plus, near-unlimited for Pro
Platform API - Free tier organization
tokens_per_minute
see https://platform.openai.com/account/limits
Platform API - Tier 1 to Tier 5 organization/model
requests_per_minute
see https://platform.openai.com/account/limits
Platform API - Tier 1 to Tier 5 organization/model
tokens_per_minute
see https://platform.openai.com/account/limits
Platform API - daily token cap organization/model
tokens_per_day
see https://platform.openai.com/account/limits

Policies

Tier progression
API usage tiers (Free, 1-5) auto-promote based on payment history and cumulative spend; higher tiers carry significantly higher RPM/TPM ceilings.
Backoff Strategy
Use exponential backoff on 429 and honor x-ratelimit-reset-* hints. The OpenAI client libraries implement automatic retries with jitter.
Per-model limits
RPM/TPM caps are enforced per model; using a higher-cost reasoning model consumes a different bucket than GPT-4o.
Batch API
The /v1/batches endpoint provides a 24-hour async path with much higher token quotas and ~50% discount for non-time-sensitive workloads.

Sources