As AI agents repeat multi-step work, users need to examine included usage, limits and additional charges alongside the monthly subscription fee. This does not establish that OpenAI and Anthropic announced the end of their $20 subscriptions. The pricing commentary referenced by this article and policies announced by the companies are different kinds of information.
API prices do not establish a provider's compute costs
Pricing a workload at published API rates can produce a different figure from a monthly subscription. API rates are customer-facing prices. They do not establish the provider's underlying compute costs or subscription losses. Subscription access conditions and API billing units also differ, so a useful comparison must include both the work and its limits.
Consider a hypothetical team automating document analysis. Comparing only the cost of one question could omit retries, review and repeated reading of documents. This illustrates a comparison method; it is not a reconstruction of a company's bill or losses.
Assessing a provider's profitability requires evidence about service costs, the scope of revenue and the period measured. The difference between an API-price estimate and a subscription fee is not automatically a loss. Nor should a few heavy users be treated as representative of all subscribers.
Usage billing can coexist with a monthly subscription
GitHub's April 27, 2026 announcement described a June 1 transition from premium request counts to AI Credits based on token consumption. It retained base monthly subscription prices while separating included and additional usage. Describing this as the abolition of monthly subscriptions is inaccurate.
A higher-priced tier also does not establish that a lower-priced plan is ending. OpenAI's April 9, 2026 update presented the new $100 Pro tier alongside $20 Plus. The price of an upper tier and the continued availability of an existing plan need separate verification.
These examples distinguish what a base fee includes from the terms for heavier use. They do not establish that every provider will change pricing on the same schedule or in the same way.
Compare completed work and actual charges
A practical evaluation starts with representative tasks and records total usage through completion. Include retries, review and parallel work, rather than just the first response. Multiple agents can generate more calls, but the increase is not necessarily a fixed multiple or exponential.
Next, identify where included usage ends and additional payment begins. Model access, interruptions caused by limits and human review time can affect value even when headline prices match. Tracking the cost per completed task and review time provides a clearer comparison than a price table alone.
Future price increases are scenarios to consider in a budget. Keep them distinct from confirmed policies and update assumptions when contracts or actual charges change.
Correction — September 16, 2026: We corrected a headline and text that presented the end of OpenAI and Anthropic's $20 subscriptions as established fact. We removed the treatment of API selling prices as compute costs, unverified loss and financial figures, and assertions that subscriptions would universally end. The commentary now accessible has a different site address and displayed date from our collection record, so we have not reconstructed its exact wording at the time of publication.
