GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
All resources

· reAPI Team

Do AI API credits expire? Plan for occasional use

Compare AI API credits expiry rules for fal.ai, Replicate, WaveSpeed and reAPI. Build a purchase ledger and avoid buying more than occasional projects need.

Do AI API credits expire? Plan for occasional use

A small API project can lose more money to unused credit than to an expensive model. Suppose you buy $60, use $24 on work you need, and let the remaining $36 expire. You used 40% of the purchase. Each dollar of useful credited consumption cost you $2.50 in cash committed.

Those are hypothetical numbers, not a provider benchmark. They show why an occasional user needs to ask when a balance expires before comparing per-image or per-token prices. A lower advertised rate is less helpful if most of the prepaid purchase never serves the project.

Some paid credits do expire. As reviewed on November 15, 2026, fal.ai gives purchased credits 365 days from purchase; Replicate specifies one year. WaveSpeed says its balance does not expire, and reAPI's terms say credits do not expire while the account remains active.[1][2][3][4] The credit type and the attached conditions matter as much as the provider's name.

Check what created the balance

A purchase, a trial grant and a recurring monthly allowance can all be labeled “credits.” They do not necessarily share a deadline. Start with the receipt or grant record that added the amount you are planning to use.

For fal.ai, the published 365-day rule applies to purchased credits. Free credits and coupons have grant-specific validity ranging from one week to one year.[1] A team that remembers “fal credits last a year” could therefore schedule work too late for a particular promotion.

A monthly allowance needs its own terms too. “AI credits per month” tells you that something is allocated on a recurring basis; it does not establish rollover, which operations qualify or whether a separate purchase follows the same rule. Read those conditions for the specific product instead of borrowing an answer from a paid-balance policy.

The account receiving the credit belongs in this check. Record the organization or account, the purchase reference and the date. If the balance looks wrong, confirm that you are viewing that account before calculating how much may have expired. A colleague's personal account and the team's account are not interchangeable billing views.

Nor does an API key's age tell you when a purchase expires. Key validity depends on the provider and key configuration; the credit record describes funding. A new key does not establish a new purchase date, and a funding problem does not by itself show that the key expired. Inspect authentication and billing separately when an integration stops working.

The published rules do not all mean the same thing

This comparison is a dated reading of the providers' public policies. It is not a test of a customer's account or a substitute for a special grant or agreement.

ProviderPublished ruleKeep this qualification
fal.aiPurchased credits expire after 365 daysFree credits and coupons have their own grant validity
ReplicatePurchased credit expires one year after purchaseThe documented prepaid purchase is non-refundable
WaveSpeedCredit balance does not expireTransfer and refund treatment are separate questions
reAPICredits do not expire while the account is activeThe active-account condition remains part of the rule

The sources are the fal.ai FAQ, Replicate prepaid documentation, WaveSpeed refund policy and reAPI terms.[1][2][3][4]

Non-expiring credit can be convenient for a project that pauses between releases. It still represents money committed to a service. If the team abandons the feature or needs a model the service does not provide, the absence of an expiry date does not turn the balance back into cash.

Similarly, a one-year deadline is not an instruction to buy a year's forecast upfront. For an uncertain experiment, the forecast itself may be the weakest part of the decision. Before comparing the deadline with the expected schedule, separate work already approved from features someone might build later.

Revisit the $60 purchase before making it

In the opening example, the team needed only $24 of credited work before the rest expired. The provider's unit price did not become 2.5 times higher. The team's cash outlay was 2.5 times its useful consumption because it committed more than it used.

That distinction helps answer why API work can feel expensive even when a model's listed rate is low. The job's billing unit and settings affect consumption. Retries and rejected outputs affect how much of that consumption is useful. An unused expiring purchase adds another cost to the team's outcome. Combining those effects into “the API price” makes it harder to decide what to change.

For the next purchase, begin with a named workload. It might be a fixed set of product images for a release or a small document-processing experiment. Identify the model, intended settings and outputs needed, then estimate the credit that work may consume. Include plausible retries without assuming every generated result will be acceptable.

Make a second forecast for the project being delayed or reduced. If the purchase only makes sense when every optional feature ships on time, it is a risky amount to commit to a deadline. A smaller initial purchase can let the team learn the actual consumption before funding more, subject to the provider's purchase conditions.

Smaller purchases have their own cost in attention. Someone needs to notice that the balance is low and decide whether to continue. If automatic funding is enabled, include those later purchases in the review; the first small payment no longer describes the total cash committed. The auto-recharge limits guide explains why a purchase allowance and a workload budget are different controls.

Do not generate unwanted images just to improve the utilization percentage. If the extra outputs have no use, converting expiring credit into those outputs does not rescue the purchase. Advance real work when it makes sense, or accept what the first experiment taught you and reduce the next commitment.

Give each purchase a date and a purpose

For a handful of purchases, a shared spreadsheet is enough. Keep one row per purchase or grant, with its account, amount, reference, purchase date, deadline or continuing condition, and the work it was intended to fund. Add the person responsible for reviewing it.

Why separate rows? Imagine two purchases made several months apart. A combined dashboard balance can be correct while revealing little about which purchase remains exposed to the earlier deadline. A single “last topped up” date is not sufficient evidence that every remaining credit shares that date.

Keep the terms that applied when each purchase was made, too. If a later purchase displays a different deadline or condition, retain both records and clarify which one governs the older credit. Replacing the older row with today's advertised policy can hide the purchase that actually needs attention.

Do not invent a consumption order to fill the gap. OpenAI's prepaid guidance explicitly says available free credits are used before purchased credits.[5] That is a rule for that service, not proof that another provider consumes the oldest purchase first. If the order affects your plan and the dashboard does not make it clear, ask about the particular balance.

OpenAI says new API accounts use prepaid billing. Do not assume that a new account has a free grant. Check what is actually present in the funded organization and the terms attached to it.[5] A possible promotion should not be entered as guaranteed funding for a project.

Set a reminder early enough to change the purchase plan or schedule useful work. At the reminder, compare the remaining approved workload with the credit at risk. If the platform only exposes a combined balance, label your allocation across purchases as uncertain rather than recording a precise remainder you cannot verify. Keep the receipt and the policy evidence together for any later support discussion.

If the balance has already disappeared

First establish the movement. Compare the funded account, credit type, receipt date, displayed expiry and intervening usage. “There used to be more credit here” does not identify which of those changed.

A payment adjustment can also alter the balance. OpenAI's prepaid guidance says earlier negative balances are deducted from a new purchase, and refunded or disputed payments remove an equivalent credit amount.[5] These are explanations to check within that provider's rules, not causes to assign to someone else's account without records.

Ask support to identify the specific movement: usage, expiry, payment adjustment or another account event. Provide the receipt reference, account, relevant dates and the amount you cannot reconcile. Keep keys and full card details out of the message. The goal is to locate the change before asking what can be done about it.

If expiry is confirmed, read the applicable purchase terms and ask about the available options without assuming restoration. Replicate's public prepaid policy describes purchases as non-refundable.[2] WaveSpeed describes a support-mediated credit-transfer process, but says account data, usage history and generated content are not transferred with it.[3] A transfer option is not a general cash-refund right.

A paused project deserves another check before funding resumes. In a hypothetical product-image project, the launch might have moved and the required images might now be different. Confirm that the next purchase still serves that revised work, rather than repeating the old purchase because it once seemed reasonable. Check the intended model in the catalog, the outputs needed and the new schedule. A non-expiring balance can make scheduling easier, but it cannot tell you whether the team still needs the feature. Record the changed purpose beside the next purchase so another teammate can understand why it was made.

For the team in the opening example, the next decision is concrete: which approved work would use the next purchase, and when? If the answer is still only $24 of likely consumption, a larger balance needs another reason. The fact that the previous $36 expired is a reason to change the buying decision, not to invent a new workload to justify it.

References

  1. fal.ai. Model APIs FAQ: credit validity. Retrieved November 15, 2026.
  2. Replicate. Prepaid credit. Retrieved November 15, 2026.
  3. WaveSpeedAI. Refund Policy: credits policy and transfers. Retrieved November 15, 2026.
  4. reAPI. Terms of Service. Read November 15, 2026.
  5. OpenAI Help Center. Setting up and managing prepaid API billing. Retrieved November 15, 2026.
All resources