Transcript API vs Summary API: Which One Does Your App Need? (2026)

TL;DR: A transcript API gives you the text of a video, and you still need a second call to an LLM to turn it into a summary. A summary API does both in one request. Pick a transcript API when you need the exact words (search, subtitles, quoting). Pick a summary API when you need the point of a video, article or link (digests, notes, cards, alerts). The PageSum API is a summary API: one POST request with a link returns a summary and a TL;DR, takes 1 credit for an article or a captioned video, and handles videos with no captions by transcribing the speech. Prices below were read from each vendor's own pages on 13 November 2026 and change often.

We make PageSum, so weigh this accordingly. Every competitor fact below comes from the vendor's own pricing or docs page; where we could not confirm something, we say so.

What is the difference between a transcript API and a summary API?

A transcript API takes a video URL and returns the spoken words, usually as timed chunks. A summary API takes a URL (or raw text) and returns a condensed version of the content. Behind the scenes a summary API still needs a transcript, so the real question is how many steps you want to own.

If you only need the words, a transcript API is cheaper and simpler. If you were going to feed the transcript to a language model anyway, a summary API saves you a prompt, a chunking strategy and a second invoice.

Why not just use the YouTube Data API?

The official route does not work for other people's videos. captions.download costs 200 quota units per call and needs OAuth with edit permission on the video, so it returns a 403 for videos you don't own (YouTube Data API reference). The default quota is 10,000 units a day per project, which is 50 downloads even if you did own the videos (quota costs). It also covers YouTube only, and only videos that already have a caption track.

That is why most apps use a third-party transcript or summary API for public videos.

How do the options compare?

All facts in this table come from each vendor's own page, read on 13 November 2026. "Unconfirmed" means we could not verify it on the vendor's page.

Two things stand out. Transcript-only services are cheaper per call, because they do less. And, in the pages we read, none of them returns a finished summary from a single link: Supadata's closest feature returns structured JSON from a prompt you write, and Firecrawl does not handle video.

What does a summary API call look like?

One request, one response. No second call.

curl

curl https://pagesum.app/api/v1/summarize \
  -H "Authorization: Bearer psk_…" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://www.youtube.com/watch?v=…", "mode": "quick"}'

Python

import requests

resp = requests.post(
    "https://pagesum.app/api/v1/summarize",
    headers={"Authorization": "Bearer psk_…"},
    json={"url": "https://youtu.be/…", "mode": "key-points", "language": "en"},
    timeout=120,
)
resp.raise_for_status()
data = resp.json()
print(data["tldr"])
print("credits left:", data["credits"]["balance"])

JavaScript

const res = await fetch("https://pagesum.app/api/v1/summarize", {
  method: "POST",
  headers: {
    Authorization: "Bearer psk_…",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({ url: "https://www.tiktok.com/@…/video/…", mode: "quick" }),
});
if (!res.ok) throw new Error((await res.json()).message);
const { summary, tldr, source, credits } = await res.json();

The response includes title, summary, tldr, a source object (video duration, whether captions were used, whether the video was cut) and a credits object with what the call charged and your balance. Full parameters are in the API docs.

What happens when a video has no captions?

This is where most pipelines break, and it is the question developers ask most. Captions exist for many YouTube videos but for far fewer TikToks, Reels and Facebook videos. A transcript API that only reads captions returns an error. The options are:

1. Fail and tell the user. Cheapest, worst experience.
2. Run speech-to-text yourself. Pay a speech vendor and manage audio download, file size limits and retries. Per-hour list prices for speech-to-text were roughly 0.15 to 0.26 (USD or EUR) for the vendors we checked (AssemblyAI, Deepgram, Gladia), before you add the download and summarizing steps.
3. Use an API that falls back to transcription for you. Supadata and PageSum both do.

PageSum reads captions first and transcribes the audio only when there are none. A captioned video up to 25 minutes costs 1 credit. A caption-less video costs 4 credits for the first 10 minutes and 2 more for each further 10, up to 25 minutes. Longer caption-less videos are cut to their first 25 minutes and the response says so (source.cut is true).

How much does a summary cost per video?

Costs depend on the content, not just the call.

Plans are $9 a month for 100 credits (Starter), $29 for 400 (Builder) and $79 for 1,500 (Scale). Credits reset each month and do not roll over, Builder and Scale can buy extra credits at their plan rate, and failed requests cost nothing. Compare that with the do-it-yourself total for the same video: a transcript call, an LLM call with enough tokens for the transcript, retries, and your own queue.

How do you handle errors and rate limits?

Plan for four cases. Errors return JSON with message and data.code.

• 401 invalid_api_key: the key is missing, unknown or revoked.
• 402 insufficient_credits: the balance is lower than the most the request could cost. The response includes data.balance and data.needed.
• 422 media_too_long or video_no_captions: this video can't be processed. Show the user why instead of retrying.
• 429 rate_limited: wait for the number of seconds in the Retry-After header, then retry.

Rate limits depend on the plan: 60, 120 or 300 requests a minute, with a separate and lower cap for video requests, because they take longer. Set a client timeout of a couple of minutes for videos and keep requests for articles on a shorter one.

Which should you choose?

Choose a transcript API if:

• You need exact wording, timestamps or subtitles.
• You index or search video content yourself.
• You already run your own summarizing prompts and pay for the LLM.

Choose a summary API if:

• You show users the point of a link: a feed digest, a "read later" app, a newsletter, a study tool.
• You accept links from TikTok, Instagram or Facebook, where captions are often missing.
• You'd rather not own prompts, chunking and retries.

Many products need both. In that case, call the summary API for the user-facing text and a transcript API only where the exact words matter.

FAQ

Is a summary the same as a transcript?

No. A transcript is every word that was said. A summary is a short version of the meaning. You can get a transcript and write your own summary from it, but that takes a second call to a language model.

Does a summary API work on TikTok, Instagram and Facebook videos?

PageSum's does: it accepts links from TikTok, Instagram, Facebook, X and YouTube, plus articles and plain text up to 200,000 characters. Private videos and posts behind a login can't be fetched.

What if the video has no captions?

PageSum transcribes the speech and then summarizes it, up to 25 minutes per request. Videos that contain only music and no speech produce no usable text.

Can I choose the summary style and language?

Yes. mode takes auto, quick, key-points, detailed, eli5, eli10 or one-liner. length takes short, medium or long. language takes a code such as en or es; by default the summary is in the content's own language.

Does the API save anything to my history?

No. API requests are not saved to a PageSum account's history.

How do I try it?

Read the API docs for every parameter, see the plans, then create a key on your account page. If you just want to summarize one video by hand, paste the link into PageSum instead.

See the PageSum API →