2026-09-30 · VideoFreeTier

Veo 3.1 Free: What a Version Number Does to a Free-Tier Answer

We have not measured Veo 3.1, so this page quotes no allowance figure for it. What the dated data on this site does show is a pattern worth more than a number: when a version number moves, what changes is the capability list, not the free-tier list. The only row in our dataset that names two versions records a clip length going from 15 seconds to 30 seconds — and says nothing at all about how much is free.

Nothing on this site requires a sign-up. Every dated figure we hold is published as CSV, JSON and Markdown on the [data page](/data/) — no account, no email gate, nothing to register for before you can read a number.

A version number is the shortest-lived label you can attach to an answer

Put a version into a search box and you have made your answer expire on someone else's release schedule. "Veo 3.1" is a label that will be superseded; the question behind it — can I use this without paying, and how much — does not get superseded with it. The label moves, the question stays, and the answer you found last month now describes an object with a different name.

That is not a reason to avoid versioned searches. It is a reason to write the answer down in a form that survives the label changing: not "Veo 3.1 gives you N", but "on date D, by channel C, the free grant was V, in unit U, refilling every R". The first form dies on the next release. The second form only needs one field updated.

This site's own dataset is written in the second form, and it is worth showing why.

The only row that compares two versions changed a capability

Our dataset holds 22 dated observations. Exactly one of them mentions two version numbers in the same line, and it is this one:

Read what that row does and does not say. The vendor, on a dated official statement, announced a doubling of the maximum clip length between two versions — 15 seconds in 2.0, 30 seconds in 2.5. That is a real, checkable capability change, and it is the single most concrete thing a version bump has ever handed this dataset.

It is also completely silent on the free tier. Same row, same vendor, same date: no credit count, no daily cap, no statement about what existing free users keep or lose.

That is the pattern to carry into a "Veo 3.1 free" search. A launch post for a new version tells you what the model can now do. Whether it also tells you what the free tier now is — and whether an existing balance survives the upgrade — is a separate question, and the honest answer is that it usually goes unanswered at launch.

Six allowance rows, zero of them about a version

The dataset does contain free-tier numbers. Six rows are allowance observations, and here they are in full:

Every row's model field reads "Seedance 2.5". Not one of these six rows has a metric, value or note that mentions a version transition. No row says what the allowance was before 2.5, whether it changed at 2.5, or whether anyone's balance was reset when the version moved.

So the count is the finding: six allowance observations, zero version-bound allowance observations. Across a couple of months of dated watching, the capability table got rewritten by a version bump and the free-tier table did not visibly move with it. That is an observation about one model on this site, not a law about all models — but it is the reason we do not assume a version number implies a new free allowance, in either direction.

What "free" had to mean before it could be counted

Those six rows are six rows because the definition was fixed first, and the definition excluded more than it included. Two rows in the table above are grants this site does not treat as a baseline free allowance:

• Lovart, up to 20 generations (2026-08-11) — the note says it is *tied to an annual subscription*. A bonus you receive because you paid for a year is not a free tier; it is a paid-plan perk. It stays in the dataset because it is real and dated, but it is not what someone means by "free".
• Doubao Work, 30 days Standard membership (2026-08-25) — the grant is membership *duration*, not generations, and the note records two conditions: a desktop client and a phone-number login. The inner quota is marked unverified. Duration of access and quantity of output are different currencies, and only one of them answers "how much can I make".

Fix that boundary before you collect numbers, because it decides the count. Include subscription perks and access windows and a model looks generous. Exclude them and you are looking at what a person who has paid nothing can actually render today.

Four steps for a versioned number

When you find an allowance figure attached to a version label, this is the order that keeps it usable:

1. Write down the version the product showed you, not the version in the article. The label in the search query, the label on the page you found, and the label in the app you open can be three different strings. The one that governs your balance is the one in the app.
2. Ask whether the grant resets, carries over, or is replaced at a version change. This is the question the launch material does not answer, and it is the one that decides whether a number you found still applies to you. If nobody states it, the answer is unknown — not "probably carries over".
3. Record whether you saw the number inside the product or in writing about the product. An allowance shown in your own account is a different grade of evidence from one quoted in a roundup, and only the first one is about your account.
4. Add today's date and the channel. A number without both is the one row type this dataset does not contain, because it could not be checked or superseded later.

[Sora 2 Free](/posts/sora-2-free/) covers filling in the remaining fields for a first measurement of any model; [Kling AI Free](/posts/kling-ai-free/) covers grading who said the number before you use it.

Why this page is about 3.1 and not 3

One disclosure, because it explains the shape of this site. We made a standing decision not to publish a page for the unversioned "veo 3 free": that query's results are held by the vendor's own pages, and an independent page would have nowhere to land. The versioned query is a different query with a different result page, so it gets its own page here.

That distinction is the same discipline as the rest of this page. A version suffix is not decoration on a keyword — it selects a different object, a different set of results, and in this case a page we can honestly fill.

Scope of what this page says

Every dated figure above is an observation of Seedance 2.5 across Jimeng, Doubao, Doubao Work, Mina Labs, Lovart, BytePlus ModelArk and vendor material between July and August 2026. None of them describes Veo 3.1, and none is an industry-wide finding. The "six allowance rows, zero version-bound" count is a count of this site's own rows — it is not a claim about how often version bumps change free tiers in general.

Where this page says a question goes unanswered, that means unanswered in this dataset. It does not mean unanswered elsewhere, and it is not a number we estimated our way around.

The full dataset — 22 rows, each carrying channel, observation date and source type — is on the [data page](/data/) in three formats, free to use under CC BY 4.0 with attribution. To turn an allowance and a usable rate into an attempt count, the [attempts-needed calculator](/tools/free-tier-calculator) uses only figures from that dataset and estimates nothing on your behalf.