TPM
Tokens Per Minute, one of the core dimensions of LLM API rate limiting. It measures the total number of input and output tokens processed by all concurrent requests within a rolling 60-second window.
Timeline (last 90 days)
Signature migration is more difficult than key exchange, for reasons including size issues, trust anchors, long-lived signature artifacts and hardware limitations.
Rate limiting (429) on free LLM APIs is usually metered simultaneously across four dimensions—RPM, TPM, RPD and concurrency—and different dimensions require different remediation strategies.
RPM and TPM are two orthogonal dimensions of model API rate limiting: RPM constrains call frequency, independent of request body size; TPM constrains the total number of Tokens consumed per unit of time, and both take effect simultaneously, with a 429 returned as soon as either is hit first.
Gemini enforces RPM, TPM and daily RPD quotas, and paid tiers also have a rolling 10-minute consumption limit.
The OpenAI API measures RPM and combined TPM by model series, and on each request checks whether the estimated prompt tokens plus the requested max_tokens fall within the remaining TPM pool.
Modern LLM rate limiting is a multi-dimensional throughput constraint that evaluates RPM, TPM, RPD and TPD/consumption limits simultaneously, returning HTTP 429 as soon as any dimension hits its ceiling.
An HSM is a standalone dedicated security chip, commonly found in phones' secure elements, laptops' TPM modules, and external hardware keys.
Section 1201 of the DMCA prohibits circumventing effective technological protection measures (TPMs), and console manufacturers' encryption and signing mechanisms are generally recognized as TPMs.
AI API providers typically set limits along two dimensions: RPM (requests per minute) and TPM (total tokens processed per minute).
Mainstream model providers set quota limits by TPM or RPM, returning a 429 error when the limit is exceeded.
All Facts (10)
Signature migration is more difficult than key exchange, for reasons including size issues, trust anchors, long-lived signature artifacts and hardware limitations.
50%UnverifiedRate limiting (429) on free LLM APIs is usually metered simultaneously across four dimensions—RPM, TPM, RPD and concurrency—and different dimensions require different remediation strategies.
50%UnverifiedRPM and TPM are two orthogonal dimensions of model API rate limiting: RPM constrains call frequency, independent of request body size; TPM constrains the total number of Tokens consumed per unit of time, and both take effect simultaneously, with a 429 returned as soon as either is hit first.
50%UnverifiedGemini enforces RPM, TPM and daily RPD quotas, and paid tiers also have a rolling 10-minute consumption limit.
50%UnverifiedThe OpenAI API measures RPM and combined TPM by model series, and on each request checks whether the estimated prompt tokens plus the requested max_tokens fall within the remaining TPM pool.
50%UnverifiedModern LLM rate limiting is a multi-dimensional throughput constraint that evaluates RPM, TPM, RPD and TPD/consumption limits simultaneously, returning HTTP 429 as soon as any dimension hits its ceiling.
50%UnverifiedAn HSM is a standalone dedicated security chip, commonly found in phones' secure elements, laptops' TPM modules, and external hardware keys.
50%UnverifiedSection 1201 of the DMCA prohibits circumventing effective technological protection measures (TPMs), and console manufacturers' encryption and signing mechanisms are generally recognized as TPMs.
50%UnverifiedAI API providers typically set limits along two dimensions: RPM (requests per minute) and TPM (total tokens processed per minute).
50%UnverifiedMainstream model providers set quota limits by TPM or RPM, returning a 429 error when the limit is exceeded.
50%