Pricing glossary
A shared definition layer for the terms reused across home, compare, calculator, and model detail pages.
Core pricing terms
These definitions show up most often in ranking, compare, and single-model reads.
Input price
The normalized cost for prompt tokens, shown per 1M tokens. Prompt-heavy or retrieval-heavy workloads usually feel this first.
Output price
The normalized cost for generated tokens, shown per 1M tokens. Long-form generation often leans on this more heavily.
Cached pricing
A discounted input rate for the cached share of reusable prompt context. It is not the same thing as batch mode.
Batch discount
A provider-listed reduced execution rate for asynchronous or batch-style processing. It should not be read as realtime pricing.
Blend
A ranking shortcut equal to 75% input price plus 25% output price. It helps scanning, but it is not a substitute for your real workload.
Budget fit
The maximum monthly request count affordable under the exact request shape and budget shown on the page.
Currency and calculator terms
These terms matter most once prices are translated into workload cost.
Source currency
The currency attached to the stored provider snapshot. This is the native basis of the captured price.
Display currency
The requested viewing currency in calculator results. If FX conversion is unavailable, rows fall back to source currency.
Cache hit ratio
The share of input tokens assumed to qualify for cached pricing. It only affects the input side.
1k requests cost
A request-level planning shortcut that turns the current request shape into a comparable per-thousand-requests cost.
Source and history terms
These terms explain why a number should be trusted and how price tracking evolves over time.
Updated
The freshest primary signal. It tells you when the latest stored pricing snapshot was captured.
Checked
A secondary verification timestamp for the stored snapshot. It can match the latest snapshot when no separate verification pass exists.
Official source
The snapshot came from the active official public pricing source for that provider.
Fallback source
The snapshot came from a public fallback source because the preferred official path is not consistently crawlable right now.
Manual baseline
A manually verified bootstrap reference point used until fresher crawls replace it.
History window
A 7d, 30d, or 90d comparison against the latest point, only shown when enough stored history exists for that window.
Use with compare
Compare is best when the workload is fixed and the terms above need to stay consistent across multiple candidate rows.
Use with calculator
Calculator is where blend stops being enough and real workload assumptions start deciding monthly cost.
Use with model detail
Model detail is where source trust, history depth, and change tracking matter most before a number is used in a real decision.