OpenAI APIs · Rate Limits

Openai Apis Rate Limits

OpenAI applies rate limits per-organization and per-model across five usage dimensions — RPM (requests per minute), RPD (requests per day), TPM (tokens per minute), TPD (tokens per day), and IPM (images per minute). Limits scale with a usage-tier system (Free, Tier 1 through Tier 5) that auto-promotes accounts as cumulative spend and account age increase. Reasoning models (o-series) and mini variants have separate limit pools; check the live console under Settings > Limits for exact per-model values.

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

It captures 7 rate-limit definitions, measuring requests_per_minute, varies, and images_per_minute.

The profile also includes 6 backoff/retry policies defined and response codes documented for throttled, quotaExceeded, serverError, and serviceUnavailable.

Tagged areas include Artificial Intelligence, Language Models, and Rate Limiting.

7 Limits Throttle: 429 Quota: 429
Artificial IntelligenceLanguage ModelsRate Limiting

Limits

Free tier (GPT-4o-class, evaluation) organization
requests_per_minute · minute
3
Free trial credits; severely capped. Also limited to 200 RPD.
Tier 1 (GPT-4o), after $5+ cumulative spend organization/model
requests_per_minute · minute
500
Roughly 30,000 TPM on GPT-4o; 10x higher TPM on mini variants.
Tier 1 (GPT-5.4 / GPT-5.5) organization/model
requests_per_minute · minute
1000
Approximately 500K TPM at Tier 1 for GPT-5 family.
Tier 4 (typical scale customer) organization/model
requests_per_minute · minute
10000
Approximately 30M TPM aggregated across flagship text models.
Tier 5 (highest, $1,000+ spend, 30+ days) organization/model
requests_per_minute · minute
30000
Up to 1B input TPM on mini-class models; ~200x Tier 1.
Reasoning models (o-series) — separate pool organization/model
varies
see platform.openai.com/account/limits
o1, o3, o4-mini have their own limit pool reflecting longer compute time per request.
Image generation (gpt-image-2 / DALL-E) organization
images_per_minute
see per-tier limits in console
IPM scales with tier; varies by model.

Policies

Tier auto-promotion
Accounts auto-promote to higher tiers based on cumulative spend (e.g. $5 unlocks Tier 1, $50 Tier 2, $1,000 Tier 5) and minimum days since first payment.
Backoff
On 429, honor x-ratelimit-reset-* hints and back off exponentially with jitter. The Retry-After header is set on some 429 responses.
Token-based limits
TPM and TPD constrain combined input + output tokens. Streaming responses still count against TPM at completion.
Batch API has separate pool
Batch jobs have a separate enqueued-tokens limit and do not consume RPM/TPM of the synchronous pool.
Provisioned Throughput
Enterprise customers can purchase Provisioned Throughput Units (PTUs) for dedicated, committed capacity outside the shared tier pool.
Soft vs hard
Most limits are soft — repeated 429s do not ban the key. Sustained abuse can trigger organization-level blocks.

Sources