2026-10-22 · VideoFreeTier

Pollo AI Free: One Parallel Task, and a First Request That Never Joins a Queue

Pollo publishes exactly one number for how much its free tier will run at once: one. On pollo.ai/pricing, read on 22 October 2026 with no account signed in, the "Parallel Tasks" row reads 1 for Free, 2 for Lite, 3 for Pro and 6 for Ultra, and the Free column carries 20 credits. That is a concurrency ceiling, not a queue length, and the page gives no queue length anywhere. Pressing submit on the video composer without an account did send a real generation request — twice — and both came back HTTP 403 with the code CREDITS_INSUFFICIENT. No job was created, so nothing was ever queued, and the ceiling of one stayed a statement rather than an observation.

This page asks one question: on the free tier, how many generations can wait in line at once. Not how many credits, not how good the clips are, not which models are included. One run on one Windows machine, 22 October 2026, no account signed in anywhere.

The one number the page gives

The compare table on the pricing page carries a row called "Parallel Tasks" in its Video section. Read across the four columns on 22 October 2026, it says 1, 2, 3, 6.

The three paid tiers repeat the same figures on their own cards — "2 parallel tasks", "3 parallel tasks", "6 parallel tasks" — which is how the row can be read with confidence: three of the four numbers appear twice and agree, and the fourth is the one left over. Free has no card of its own. It exists only as the leftmost column of that table.

The label Pollo uses for the row is a sentence, not a number. It ships in the page's own script bundle as "Process up to {n} generation or enhancing tasks at the same time", and the table fills {n} with the figures above.

A ceiling is not a queue

Two different things get called "how many at once", and only one of them is published.

A concurrency ceiling says how many jobs may be running at any instant. A queue length says how many may be waiting behind them. A tool with a ceiling of one can still take a second request — it can hold it, or refuse it. Which one it does is the difference between waiting four minutes and being told to come back tomorrow.

Pollo publishes the ceiling. It publishes no queue length, no waiting-slot count and no wait estimate — not on the pricing page, not in the FAQ, not in the composer.

Three sentences for the same limit

The page ships three different descriptions of what happens when the ceiling is reached, and they do not describe the same outcome.

"You've exceeded the simultaneous task limit allowed for your plan. Your request will be processed once there's available capacity. To increase this limit, please upgrade your plan."

That one says the request is kept and handled later. That is a queue.

"Parallel task limit reached. Please wait for some tasks to complete."

That one tells you to wait, which reads as a refusal of this moment.

"You cannot generate more than one video at the same time."

That one says the second video is not something you can ask for at all.

All three are in the same script bundle on the same day, next to the plan-table row. None of them appeared in this run, because no job was created that could reach the limit. Which one a free account actually meets is the part the page leaves open — and the three answers lead to three different user experiences.

Where the word queue does appear

Queue is not a word Pollo avoids. It uses it several times, just never about the free tier.

• "Due to high demand, your request is in queue."
• "Your request has been queued up due to increased demand right now."
• "Queued… 50%"
• "You are now in the queue with reduced wait time, exclusive to paid plan users."
• "Generation tasks for Unlimited models run on a dedicated processing queue, which is independent of the 'priority queue' that consumes credits."

Every one of those belongs to peak demand, to paid priority, or to the Unlimited models a subscriber gets. The free tier's row is a number with one sentence about how many run at once, and nothing about what becomes of the rest.

What the button does without an account

The composer at pollo.ai/ai-video is open to a visitor with no account: a text box, a model row reading Text/Image to Video · Pollo 2.5 · 5s / 720p / 16:9, and a submit button. The page quietly signs the visitor in as an anonymous user before anything is pressed, and issues that session 20 new-user credits.

A prompt was typed and the button pressed. The page did not merely raise a wall — it sent the request first:

POST /api/trpc/video.createVideoByTextV2 → 403

{"message":"User video limit error","code":-32003,"data":{"code":"FORBIDDEN","httpStatus":403,"errorCode":"CREDITS_INSUFFICIENT"}}

Then it raised a full-screen panel: "Sign up now and get 20 free credits", with Google, Apple, Facebook, Kakao and email as the ways in.

The panel was closed and the button pressed again. A second generation request went out, and came back with the same 403 and the same code. A third press sent nothing at all, because the panel was up again and the click landed on it.

So the sequence is: request, refusal, signup offer. The refusal is a credit check, not a queue position — the error code names insufficient credits, not a full queue. Nothing was created, nothing waited, and the "one parallel task" limit was never the thing that stopped anything.

One job's worth of credits

That refusal is worth one paragraph of its own, because it explains why the queue question is hard to reach even with an account.

The plan table prices Pollo 2.5 at 720p at 5 credits per 5 seconds. The composer's own estimate endpoint, asked for exactly those settings on the same day, answered 12. Both numbers are pollo.ai's, both were read on 22 October 2026, and they disagree; this page keeps both rather than picking one.

Take the higher figure and the free allowance of 20 credits covers one 5-second clip at that price, with change left over and not enough for a second. A ceiling of one parallel task and an allowance that buys one job point the same way, which is a tidy result — and also the reason a free tier cannot be used to demonstrate what happens to a second request. There is no second request to demonstrate.

What was not measured

The fate of a second job on a free account: whether it would sit in a queue or be refused outright. That needs an account, and none was created for this run — signing up is an account action, and this page only reports what it could see from outside one.

Two smaller limits, both stated plainly. The pricing page rendered its paid prices in Korean won on this network, because the machine's address geolocates there; the free column and the parallel-task row are not prices and are unaffected. And the "Unlimited" declaration panel behind the page's "Learn more" link opened once and then would not reopen across three further attempts, so the wording quoted above comes from the page's own strings rather than from a second reading of that panel.

What would settle the question: a free account, two submits inside one minute, and a look at whether the second job appears as "Queued…" or as an error. Either answer is a fact. The page currently supports both.

Where this run is kept

The compare table, both 403 responses with their bodies, the composer before and after each press, the pricing page's HTML and the bundle strings quoted above are dated and stored together, so every number here can be traced to the run that produced it. Nothing on this page is an average or an estimate; it is one afternoon's reading, and it will be re-run rather than revised if the page changes.

The site's ledger of free-tier observations — each figure with the channel it was seen on, the date it was observed, and whether it came from a vendor statement, a report or a single observation — is published as CSV, JSON and Markdown on the data page.