Livepeer Agent · Usage & cost

What did I spend, and on what?

You've been generating images, video and voice-overs. Now you want the receipt: how much have I used, which models did it, and how was that price worked out? You can ask for all of it in plain English, and this page explains every line you get back.

The short version: everything is priced per unit of what you actually made — a second of video, a megapixel of image, a thousand characters of speech. Nothing else. No subscriptions, no minimums, no surprises.

One idea, and the rest follows

Every model has a price per unit. A unit is a natural measure of the thing you asked for — a second of video, a megapixel of image, 1,000 characters of speech, one image, one 3D mesh.

Your cost for a job = unit price × how many units you asked for.

Start1 · Just ask — it's one sentence

You don't need to find a dashboard or an invoice page. In Claude (Cowork, Desktop, Code) or Cursor, with the Livepeer Agent connected, type any of these:

"How much have I spent in the last 14 days?" "Show me my usage — what did I run and what did it cost?" "What's my spend this week, broken down by model?"

Behind the scenes that's the me_usage report — your own activity, under your own key. It needs no special permission, because it's your data. If you'd rather see it job-by-job rather than totalled, ask for "my cost report, per job".

Nobody else's numbers, ever. These reports are scoped to your key and cannot show another person's activity — not by asking differently, not by guessing an identifier. If you share one key across a team, you'll see the team's combined total; give people separate keys if you want separate reports.

The receipt2 · Read your report

Here's a real one. Three kinds of work — a couple of short videos, one image, three voice lines:

Your usage — last 14d Total: 6 calls (6 ok / 0 failed) · estimated cost $0.6607 pixverse-i2v 2 calls $0.6300 flux-dev 1 calls $0.0263 gemini-tts 3 calls $0.0044 Cost is an estimate at list price, not an invoice. It covers the calls we recorded a price for — including failed renders, which are still billed.

Read it top to bottom:

  1. The window. "last 14d" — how far back this covers. 14 days is the default because that's how far back detailed records are kept.
  2. The totals. 6 things ran, all 6 succeeded, and together they cost about 66 cents.
  3. The breakdown. One row per model, biggest spender first. Here two short videos account for 95% of the bill and everything else is rounding — which is the single most useful thing this report tells you.
The pattern to look for. Video is almost always the line that matters. Images and voice are typically fractions of a cent each; a few seconds of video can be more than a hundred images. If a bill ever looks bigger than expected, check the video row first.

3 · What each item means

You seeIt means
last 14dThe time window. You can ask for 24h, 7d or 14d. Ask for longer and you'll be told it was shortened to 14 days rather than handed a total that quietly leaves things out.
6 callsHow many separate pieces of work ran. One image is one call; one video is one call.
6 ok / 0 failedHow many finished, and how many didn't. See §6 — a failed render can still cost money, because the model still ran.
estimated costThe total at list price. Called an estimate for one honest reason: it's priced from what you asked for at the moment you asked, so it's the expected charge, not a settled invoice.
pixverse-i2v, flux-devThe model that did the work. You never have to name one — the agent picks — but the report shows which it used so you can steer next time ("use the cheaper video model").
2 calls · $0.6300How many times that model ran in the window, and what it came to in total.
unknown (sometimes)Work we recorded but couldn't attribute to a single model — usually a step that fans out to several. Rare, and it never inflates the total.

4 · How cost is calculated

There is exactly one formula, and it's the one from the top of this page:

cost = unit price × number of units

What counts as a "unit" depends on what kind of thing you're making. There are five:

Kind of workUnitCounted asReal prices today
Videoper secondThe length of the clip you asked for. A 5-second clip is 5 units.$0.0394 – $0.42 /sec
Image (some models)per imageOne picture, one unit — size doesn't change it.$0.0022 – $0.063 /image
Image (other models)per megapixelHow big the picture is. A 1024×1024 image is ≈1 megapixel.$0.0032 – $0.063 /MP
Speechper 1,000 charactersThe length of the script you gave it, counted in characters.$0.0053 – $0.026 /1k chars
3D / music / one-offsper itemOne mesh, one track, one job — a flat price per result.$0.10 – $0.21 /item

Editing steps — trimming, cropping, adding captions, stitching clips together — are priced the same way (per second of video touched) and are typically a tenth of a cent per second, so they round to nothing next to generation.

Want the exact price before you commit? Ask "what does a 6-second video cost?" or "show me pricing for image models". You'll get the per-unit price for each option, and the agent tells you the price of a job before it runs anything billable.

5 · Worked examples — the same report, line by line

Every number in the report above can be reproduced with the one formula. Here it is:

The video line — $0.6300

StepValue
Model usedpixverse-i2v (turns a still image into a moving clip)
Unit price$0.063 per second
Clip length5 seconds
One clip5 × $0.063 = $0.315
Two clips2 × $0.315 = $0.63

The image line — $0.0263

StepValue
Model usedflux-dev (text → image)
Unit price$0.02625 per megapixel
Image size1024×1024 ≈ 1 megapixel
One image1 × $0.02625 = $0.0263

The voice line — $0.0044

StepValue
Model usedgemini-tts (text → speech)
Unit price$0.01575 per 1,000 characters
Script length~93 characters per line (a short sentence)
One line0.093 × $0.01575 ≈ $0.0015
Three lines≈ $0.0044
Sense-check for the whole report: $0.63 + $0.0263 + $0.0044 = $0.6607 — the total at the top. Nothing is added on afterwards. There is no per-call fee, no storage charge, no monthly minimum.

Honest6 · Why a number may surprise you

A failed render can still cost money

If a model runs and then fails — a timeout, a refusal, a bad frame — the work was still done, so it's still charged. That's why the report shows ok and failed counts side by side: a bill bigger than your finished-output count usually means retries. If a job fails for a reason you think is ours, say so and we'll look at it.

Some calls show as calls but not as cost

A request that fails before a model runs — a bad input, an unreachable service — costs nothing, so it's counted as a call with no price. Your total is never inflated by these; it can only be a slight under-count of things that never actually ran.

"Estimate" means expected, not provisional

The price is worked out from your request at the moment you make it — the model, the length, the size. For the overwhelming majority of jobs that is the charge. It's called an estimate because occasionally a model returns slightly more or less than asked (a clip rounded to the nearest whole second, say).

The window is 14 days, on purpose

Detailed per-call records are kept for 14 days. Ask for 30 days and you'll be told it was shortened, rather than shown a "30-day total" that's really 14 days of data. If you need longer-term records, keep the reports you pull — or ask us about longer retention.

Checking on a job doesn't cost anything

Long renders are polled for progress. Those checks are free and are never double-charged — one render is charged once, however many times you or the agent look at it.

7 · Four ways to spend less

  1. Draft fast, finish well. Say "make it fast" for the first pass. The quick video model is a third the price of the premium one and returns in under a minute. When you like the take, say "now re-render that one in HQ" — you pay the premium price once, for the one you're keeping, instead of on every experiment.
  2. Keep test clips short. Video is priced per second, so a 4-second test costs a third of a 12-second one. Get the look right at 4 seconds, then extend.
  3. Ask for the plan before the render. "Plan it first, don't render yet" gets you the shot list and the price with nothing spent. Especially worth it before anything that fans out into many clips.
  4. Set a ceiling. Ask to "set my spend cap to $5 a day" and work stops before it crosses the line instead of after. Cheap insurance against a runaway loop.
Rule of thumb. A hundred draft images cost about the same as ten seconds of premium video. Iterate freely on stills; be deliberate with video.

8 · FAQ

Do I need permission from an admin to see my own spend?

No. Your own usage and your own cost are yours to read, always. (An everyone's-usage view — for whoever runs the account — is separate and is the only one that's restricted.)

Is there a subscription or a minimum?

No. You pay for units of output. Generate nothing this month and it costs nothing.

Am I charged for the agent's thinking, or just the output?

Just the output — the images, clips, audio and edits it produces for you. Planning and conversation aren't billed as media.

Why is one video model 10× another?

Quality and speed. The range runs from about 4 cents a second to 42 cents a second. Ask "what are my video options and what do they cost?" and pick per job — a social draft and a hero shot don't need the same model.

Can I get this as a document to file or share?

Yes — ask for "my full usage report" and you'll get a complete, self-contained version with every row, suitable for saving or printing.

My total looks wrong. What now?

Ask for the per-job cost report rather than the model totals — it lists each job with its model and price, which usually makes the culprit obvious (nearly always a longer-than-expected video, or retries). If it still doesn't add up, send us the report and we'll trace it.