Video transcoding cost is per minute. A ladder bills five times.

Video transcoding cost is quoted per minute of output, not per minute of input, so one ten-minute source rendered into five bitrates bills as fifty output minutes. On Google Cloud Transcoder that ladder runs $1.50 for the single job, and the 2160p rung alone bills four times the 360p rung at $0.060 against $0.015 per minute (Google Cloud). The rate on the pricing page is never the number on the invoice, because your rendition count is the multiplier.

That gap is why every "transcoding price per minute" comparison you find is useless for budgeting. You know how many files you process and roughly how long they are. You do not know, until you read the fine print, that resolution tiers the rate, that frame rate and codec stack on top of it, that outputs under ten seconds get billed as ten, or that storage and egress show up as their own lines after the encode finishes.

The per-output-minute meter charges once per stream encoded

Google states the rule plainly on its own pricing page: the per-minute price is calculated for each video stream encoded. Its worked example takes a 30-minute input and requests 640x480, 1280x720, 1920x1080 and 3840x2160. Those four streams bill $0.015 + $0.030 + $0.030 + $0.060, which is $0.135 per minute of input duration, for a total of $4.05 on one job. Audio-only output is $0.005 per minute, and generated subtitles are billed separately.

AWS Elemental MediaConvert works the same way with different numbers. It bills per minute of output video, with the Basic tier starting at $0.0075 per minute and Professional at $0.0120, and publicly quoted resolution tiers of $0.0075 SD, $0.0150 HD and $0.0300 UHD on the Basic AVC single-pass path (AWS). Two details on that page matter more than the headline. Each output is measured in one-second increments converted to fractional minutes, and the minimum billable duration is ten seconds, so a six-second clip bills as ten. AWS also bills in what its documentation calls normalized transcoding minutes: output minutes multiplied by the features you used, with resolution class and frame-rate band (up to 30, 60 or 120 fps) folded into the usage type at 0.01-minute increments (AWS docs).

The discount end of this market publishes rates near half a cent per output minute for base-codec SD. Treat that number as a headline rather than a rate: those same pages list HEVC and VP9 at twice the base, AV1 at three times, another 100% for each additional 30 fps, and separate line items for repackaging, audio, GIF and thumbnails. A 60 fps AV1 output is six times the advertised figure before anything else is added. Those are published list structures, not figures I can source to a neutral page, so treat them as an estimate when you model them.

One ten-minute source, priced four ways

Converting a per-minute rate into cost per finished video is the only comparison that survives contact with a real pipeline, so here is the same ten-minute file through four different requests. FFmpeg Micro meters on input: 1 token per 10 seconds, so ten minutes is 60 tokens, which is $0.48 at the Starter grant rate of $19.99 for 2,500 tokens and about $0.43 on Pro at $49.99 for 7,000 (pricing).

Read the losing rows first, because they are the honest part. For a single SD or HD output at volume, the per-output-minute meter is cheaper per encode, and nothing in a flat input rate changes that. For a genuine five-rung ladder, it is cheaper outright: each FFmpeg Micro request produces one output, so five renditions is five jobs billed against the same ten minutes of input, and five times a flat rate beats nothing.

Two rows tell the other story. At 2160p the flat input meter is already cheaper than Google's UHD rung, because resolution does not tier the rate at all: 4K costs what 360p costs. And the trim row is where input metering usually breaks against vendors who bill the whole source file. FFmpeg Micro bills the clip, so pulling 30 seconds out of a one-hour recording is 3 tokens, not 360.

What lands on the invoice after the encode

The transcode line is rarely the biggest number in a video bill. The renditions have to live somewhere and then travel to viewers, and both are metered separately from the encode. Amazon S3 Standard starts at $0.023 per GB-month for the first 50 TB, with data transfer out billed per GB above the first 100 GB a month (AWS).

Run the five-rung ladder above at 800 kbps, 1.2, 2.8, 5 and 15 Mbps and you get roughly 1.86 GB of output per ten-minute source. That is about $0.043 a month in S3 Standard for one video, which sounds like rounding error until you multiply: a thousand videos a month is 1.86 TB added, roughly $43 in month one and $43 more every month after, before a single byte is delivered. One cost breakdown of three-cloud video infrastructure puts CDN egress at 50% to 70% of a streaming bill against 10% to 20% for encode (Spendark). The encode meter everyone compares is the smallest of the three.

FFmpeg Micro has one meter and no second invoice: checking status, listing jobs and downloading are free, storage is unmetered on every plan including Free, and failed jobs are never billed. The request is the whole integration, and it looks like this from any language, n8n, Make, Zapier or an MCP-connected agent:

curl -X POST https://api.ffmpeg-micro.com/v1/transcodes \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": [{"url": "https://storage.example.com/master.mov"}],
    "outputFormat": "mp4",
    "options": [
      {"option": "-vf", "argument": "scale=1080:1920"},
      {"option": "-c:v", "argument": "libx264"},
      {"option": "-crf", "argument": "23"}
    ]
  }'

You poll that job id until it reports completed, then download. If it fails, you pay nothing and retry, which is a different risk profile from a per-job processing fee or your own encoder fleet burning machine-hours on a bad input. The failure modes worth handling in that loop are covered in why your transcode job failed.

Which meter fits the job you actually have

Pick the meter from the shape of your output, not from the headline rate. Three cases cover almost everything a builder runs into.

If you need a packaged HLS ladder with a master playlist sitting behind a CDN, buy the per-output-minute meter and stop comparing it to anything else. You are purchasing a delivery platform, the encode is 10% to 20% of what you will spend, and the vendors on that meter bundle the packaging and the origin you would otherwise build. Price it as a delivery decision. The same logic applies when you are already storing and serving from S3 or Cloud Storage and your outputs are SD or HD singles at volume, where $0.0075 to $0.030 per output minute is genuinely hard to beat.

If what you need is one file at a platform's spec, handed back to the next step in a pipeline, the flat input meter is the simpler buy and gets cheaper as resolution goes up. A 9:16 crop for Reels, a watermarked UGC clip, an H.264 normalization of whatever a user uploaded, burned-in captions: one output, one charge, no storage line, no egress line, no resolution tier. That case is worked through in normalize every upload to one house spec.

FFmpeg Micro is the wrong move in three specific situations, and they are easy to check before you sign up for anything. Direct HLS output is not in the API today, so packaged adaptive streaming needs a packaging step elsewhere. Five renditions of every file means five billed jobs, and the per-output meter wins that arithmetic. And input size is capped at 100 MB on Free, 1,024 MB on Starter and 2,048 MB on Pro, so a two-hour 1080p master has to be trimmed or split before it is a job at all. If you want hosting, a player and analytics in the same bill, the three-meter breakdown of a video platform's pricing is a better place to start than this one.

FAQ about video transcoding cost

How much does it cost to transcode a 10-minute video?

A ten-minute video transcoded to one 1080p MP4 costs $0.30 on Google Cloud Transcoder at $0.030 per output minute, about $0.15 on AWS Elemental MediaConvert's Basic tier at $0.0150, and $0.48 on FFmpeg Micro's Starter plan, where ten minutes of input is 60 tokens. Ask for five renditions instead of one and the first two numbers multiply by five while the third becomes five separate jobs.

Why does a five-rendition ladder of one file bill five times?

A five-rendition ladder bills five times because the cloud transcoders meter minutes of output, and each rendition is its own encoded stream. Google Cloud Transcoder states that the per-minute price is calculated for each video stream encoded, which is why its own example of a 30-minute input with four requested resolutions totals $4.05 rather than the $0.90 a single HD output would cost.

Is there a minimum billing increment on cloud transcoding?

AWS Elemental MediaConvert publishes a minimum billable duration of ten seconds per output, so a six-second clip is billed as ten seconds, and it measures usage in one-second increments converted to fractional minutes. FFmpeg Micro has no minimum increment and no rounding up to the minute: a ten-second video costs ten seconds of metered time.

Do I get charged for a transcode that failed?

On FFmpeg Micro, failed jobs are never billed, and you can retry without paying for the attempt. On a per-output-minute meter the answer depends on how far the job got before it failed, and on your own encoder fleet you pay for the machine-hours either way, so check the vendor's billing documentation rather than assuming a failure is free.

How do I compare a per-output-minute price to a per-input-second price?

Multiply the per-output-minute rate by the number of renditions you request and by the resolution tier each one falls into, then divide the total by one finished video to get a figure you can compare to a flat per-input rate. The crossover sits between HD and UHD: at 1080p the output meter is cheaper, at 2160p a flat input rate is, and above one output per file the output meter wins again.

Sixty tokens is what that ten-minute source costs on a flat input meter, whether it comes back as 360p or 2160p, and a new account arrives with 200 of them, roughly 33 minutes of video, no card needed. Sign up free and run your own file through the arithmetic above: no servers to run, no FFmpeg to install, called from your code, from n8n, Make and Zapier, or from your AI agents.