Name the constraint correctly

Common usage terms
TermUsually meansDo not assume
TokenA model-specific unit of encoded input or outputOne token equals one word
CreditA service-defined allowance or billing unitOne credit has the same value everywhere
Rate limitHow much activity is allowed during a time windowIt is the same as a spending cap
Usage capA plan, budget or policy ceilingUnused capacity carries forward
Context limitHow much information a model can consider in one interactionAll available space should be filled

Compare services with a common task

A plan described in credits cannot be compared directly with token-priced access until you know what actions consume a credit, whether different models consume different amounts, and which limits apply. Run the same representative task and record accepted completions per unit of spend and per unit of time.

Also record what happens at the limit. A request may queue, fail, switch model or require a higher plan. That operational effect can matter more than a small difference in nominal unit price.

Ask before committing to a plan

  • Which inputs, outputs, tools and retries are metered?
  • Are limits per user, project, model or organisation?
  • What resets, expires or rolls over?
  • Can usage be exported and attributed to a task?
  • What happens when a cap or rate limit is reached?
  • Can a budget alert or hard ceiling be set?

Keep accounting separate from value

Commercial units change. A durable measurement layer keeps your task ID, result status, latency and review outcome stable while mapping each provider's usage fields separately.

Sources & further reading

  1. Models

    OpenAI Developers. Official model-selection documentation, checked 13 September 2026. Product names, capabilities and prices can change; the collection uses the durable principle of matching capability to a task rather than prescribing a current model.

  2. GenerationConfig REST resource

    Google Cloud. Official documentation showing that output-token limits are a controllable generation setting. Parameter names and behaviour are provider-specific.

Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.

A correction, a counterexample or an experience worth sharing?

Join the conversation ↗