Frequently asked questions
Does changing a model change the table order?
On the home comparison table, selecting a model updates only that provider row. Click an input or output column arrow to sort explicitly. A low input price does not necessarily mean a low total cost.
How can AI model performance and cost be compared?
Compare benchmark scores and measured task costs within the same evaluation version, while checking each model’s reasoning settings. Input/output prices alone do not measure performance. Missing measurements are left unknown.
Is the highest-scoring model always best for me?
No. Benchmark scores describe a specific evaluation. Your task, latency, reasoning settings and total cost can change the choice. Use the score and cost columns together, and open the benchmark source for its conditions.
Where do AI news and reset records come from?
News comes from configured official publisher pages and feeds. Reset records link to reviewed official announcements. Competitor sites help discover original links; their APIs or datasets are not used as our service data.
How often is information checked, and can announcements be missed?
The collection schedule is four checks per day; a site rebuild is requested only if published information changes. A failed source keeps its last verified data. Known X posts are rechecked, but without an X API connection new X posts are not discovered automatically. Source status is shown on the methodology page.
Does receiving a reset pass automatically reset my usage?
A pass grant and a completed usage reset are different events. A delivery announcement alone does not confirm redemption or an automatic reset. Check the original announcement and the product usage page for the applicable rules.
Does a Codex reset also reset every ChatGPT limit?
Do not assume so. A record applies to the product, plan and users named in its official announcement. We keep product scopes separate and leave unspecified eligibility unknown.
Why does the elapsed-day number not change at midnight?
Elapsed days count completed 24-hour periods from the recorded event time, rather than calendar date changes. For example, 23 hours after an announcement is still day 0. The browser refreshes the count once per minute.
Can credits be used for both subscriptions and API calls?
Only the announcement can establish which product a credit grant covers. Do not treat a product-specific credit as a universal balance. Check eligibility, amount, claim requirements and expiry in the event detail and original source.
Can I assume every input token is billed at the cached price?
No. A listed cached rate does not guarantee a cache hit. Use cached billing only for input covered by the provider’s cache rules. The calculator requires cached input to be no greater than total input and does not assume a discount when its rate is unknown.