Inspect the price changes recorded by this reference and follow their official evidence. Use the history to explain an estimate change without inventing earlier observations.
Explore recorded prices
Explore recorded price changes
Your text and estimates stay in this browser. No API requests are sent to model providers.
Loading verified model records…
Read the chart with its change table
Select the provider and model relevant to your decision, then inspect the series label and unit. The chart is a visual view of recorded data; the accompanying table retains the change details and source. A step between observations describes the history available here. It does not prove that a provider changed its terms at every point represented by a line.
Read the initial observation differently from a later change. An initial stored rate establishes the first evidence in this reference, not the model’s original launch price. A missing earlier interval is missing history. The site must not backfill a plausible-looking trend from current pricing or pretend that a fetch occurred before the actual collection time.
Official pricing distinguishes categories and qualifying conditions, which must remain attached to a historical comparison. Official documentation.
Keep input, output and cached categories separate. A change in one series does not establish that all categories moved together. The same caution applies to processing modes and tier conditions. Before describing an increase or decrease, verify that the old and new records refer to the same billable work.
Separate observed time from a provider announcement
A source observation records when the reference successfully checked the evidence. A provider announcement may state when a new condition becomes effective. Those are different facts. When an effective date is documented, preserve it with the source; when it is absent, do not infer one merely from the collection schedule.
If the parser encounters an unusual change, the review process may hold the proposed update while the last accepted value remains public. A later publication time therefore need not be the first time the provider page changed. Read any accompanying note before drawing a precise chronology from the public table.
A failed source check does not establish that prices stayed the same. Inspect freshness and collection state when the history looks unexpectedly quiet. An apparently flat chart is meaningful only within the evidence actually collected, and the source inventory provides the context needed to judge that evidence.
Explain a budget change
Reopen the cost calculator with the previous workload assumptions, then compare the affected model and category. Hold the workload steady while assessing the price effect.
After isolating the reference-rate change, inspect application changes separately. Longer answers, more retrieved material, additional model stages and reduced cache reuse can increase cost even when provider rates are unchanged. Keep an old and a current workload scenario so the explanation does not attribute every invoice movement to a model-price update.
Use the history as an input to a migration review, not an automatic switch trigger. A cheaper category may improve the estimate while a different model requires substantial evaluation or feature changes. Record the expected saving, the integration work and the acceptance checks needed before deciding whether the change is worthwhile.
Follow and correct the record
The changelog groups published updates and offers a feed. The source inventory shows the underlying collection state. Use both when a decision depends on the timing or freshness of a recorded change.
For a correction, identify the exact model, price category, condition and source evidence. A general statement that a provider is cheaper is not enough to repair a historical row. If the official source has changed again, preserve a precise description of the discrepancy without inventing an archived response that was never collected.
Last verified · Source ↗
Frequently asked questions
Does the earliest point show launch pricing?
Does a quiet chart prove prices stayed unchanged?
Can I compare every line directly?
Why might publication follow observation?
How do I isolate a price change in my estimate?
What should a historical correction include?
Sources
Last verified · Source ↗