Request Budget

Plan for the busy hour

A rate limit is a ceiling. Reserved capacity is a different purchase.

01

Answer now

Groq

Developer limits
250K TPM / 1K RPM

GPT-OSS-120B Developer plan

Rate ceilings are not reserved throughput or achieved speed. Context 131,072; max completion 65,536 tokens.

Source ↗

OpenAI

GPT-6.1 Sol usage tiers
Build: 5,000 RPM / 1,000,000 TPM

gpt-6.1-sol API usage-tier limits

Launch lists 10,000 RPM / 4,000,000 TPM; Grow 15,000 RPM / 40,000,000 TPM. Free unsupported. Usage/spend can raise the tier; actual account limits and achieved throughput were not checked.

Source ↗

Together AI

Serverless
Best effort; 429 or 503 under demand

Together Serverless

Limit retries to the application latency budget. Provisioned throughput reserves selected model/family capacity under a separate SLA.

Source ↗

Amazon Bedrock

Account quotas
Per-model and per-region; account quota may be lower

Amazon Bedrock Runtime / Mantle

Embedding models may be RPM-only. Account age, payment and usage history can affect quotas; no automatic increase or common throughput promise.

Source ↗

Cerebras

GPT-OSS-120B PayGo limits
500K uncached TPM / 1K RPM

Cerebras Developer PayGo, GPT-OSS-120B

Models and limits may change without notice. Custom higher limits, SLAs and dedicated capacity require agreement; no production suitability inferred from a self-serve limit.

Source ↗
02

Let the job wait

Fireworks AI

Batch
50% off input and output

Fireworks Batch jobs

For work that does not need an immediate answer. Not a blanket claim of immediate serving at half price.

Source ↗

Anthropic

Message Batches
50% off input and output

Claude Message Batches

Batch and Fast cannot be combined. Partner regional pricing and model-specific long-context conditions can differ.

Source ↗

OpenAI

Batch and Flex
50% lower than Standard

gpt-6.1-sol API

Fast 2× Standard; Ultrafast 6×. Long-context and regional conditions still matter; subscription use has different terms.

Source ↗

Microsoft Foundry

Global Batch
50% discount; 24-hour processing

Global Batch, selected supported models

Processing may occur in any Azure region. Data Zone Batch stays within its zone. Not every model supports every deployment type.

Source ↗
03

Reserve or handle interruption

Fireworks AI

Reserved Throughput
Separate dollars-per-minute capacity offer

Fireworks Reserved Throughput

Commitment required. Vendor lists a 99.9% uptime SLA, subject to service terms. Usage above reserved capacity is charged at Standard rates; no contract remedy independently reviewed.

Source ↗

Fireworks AI

Spot
30% off Standard; throughput can fall to zero

Fireworks Spot Serverless, select models only

Retry or fall back to Standard. Not reserved capacity and not every model.

Source ↗

Microsoft Foundry

Reserved capacity
Provisioned types bill reserved PTU

Global / Data Zone / Regional Provisioned

Model availability and capacity differ by type. Developer evaluation is a separate 24-hour-lifetime type with no SLA or data residency guarantee.

Source ↗

OpenRouter

In-region routing
Business and Enterprise; US/EU

OpenRouter US/EU in-region routing

Standard has data-policy routing but not these in-region features. Contractual SLA Enterprise only; paid rate limits pass through underlying provider.

Source ↗

What a catalog cannot reserve

Your model, account, region and requested mode decide the actual limit. We have no live quota check, concurrency test or comparable latency trace.

Check the exact model and endpoint →