CapsAI

Formats

Subtitle API comparison: SRT and VTT from OpenAI, Deepgram, AssemblyAI, Google, AWS and Rev AI

Six speech-to-text APIs compared on SRT/VTT output, word timestamps, live streaming and list price (4 Nov 2026), plus OpenSubtitles and YouTube caption APIs.

A subtitle API is one of two things, and the search results mix them up. Most people want an API that generates subtitles: you send audio or video, it runs speech-to-text and returns timed text, ideally as an SRT or VTT file. The other kind fetches subtitles that already exist: OpenSubtitles for films and TV, and the YouTube Data API for caption tracks on YouTube videos.

The short answer for generation, from each vendor's own docs as read on 4 November 2026: AssemblyAI, Rev AI, Amazon Transcribe and Google Cloud Speech-to-Text (V2 batch) return SRT and VTT from the API itself. Deepgram returns word timestamps and makes SRT or WebVTT with its helper libraries. OpenAI still lists srt and vtt as response formats, but its subtitle examples use whisper-1, which shuts down on 26 February 2027. For fetching: OpenSubtitles needs an API key and caps downloads, and YouTube only lets you download caption tracks for videos you can edit.

Subtitle generation APIs compared

Every cell below comes from the vendor's own pricing page or docs, read on 4 November 2026. Prices are copied as each page shows them, in US dollars.

VendorSRT / VTT outputWord timestampsStreaming (live)Languages, as statedList priceFree tier
OpenAIsrt, vtt listed; gpt-4o-transcribe models return json onlywhisper-1 onlyYes: gpt-live-transcribeWhisper: 98gpt-transcribe $0.0045 / minute; Whisper $0.006 / minuteNot verified
DeepgramVia helper library, not the APIYesYesPer-language table, no totalNova-3 pre-recorded $0.0043 / min$200 credit
AssemblyAIYes: srt or vtt endpointYes (ms)YesUniversal-2: 99; Universal-3.5 Pro: 18Universal-2 $0.15/hr; Universal-3.5 Pro $0.21/hr$50 in credits
Google Cloud Speech-to-TextYes: V2 batch, to Cloud StorageYesYesPer-model table, no totalV2 $0.016 / minute; dynamic batch $0.003 / minute60 min a month (V1 table)
Amazon TranscribeYes: batch jobsYesYes, not all languagesTable of languages and dialects$0.006 per minute batch (US East example)60 min a month for 12 months
Rev AIYes: captions endpointNot verifiedYesReverb: English; foreign: Spanish, French, Chinese, Portuguese and 53 moreReverb $0.20 per hour5 hours of Reverb

What one hour costs at list price

Our arithmetic from the prices above, for one hour of English audio, before any free credit:

Vendor and modelRecorded hourLive (streaming) hour
AssemblyAI Universal-2 / Universal-Streaming$0.15$0.15, billed while the socket is open
Rev AI Reverb$0.20Not listed
Deepgram Nova-3 monolingual$0.26$0.46
OpenAI gpt-transcribe / gpt-live-transcribe$0.27$1.02
Amazon Transcribe (US East example rates)$0.36$0.60
Google V2 standard / dynamic batch$0.96 / $0.18Not listed separately

Per-hour prices are close enough that format and workflow usually decide, not cost. A 10-minute video costs cents with any of them.

Vendor notes that the table cannot hold

OpenAI. The API reference lists srt and vtt among response formats, then says gpt-4o-transcribe and gpt-4o-mini-transcribe support json only. Word and segment timestamps need verbose_json and, per the guide, work only on whisper-1. The guide marks whisper-1 and the gpt-4o transcription models as deprecated, shutting down on 26 February 2027, and recommends gpt-transcribe, which returns the transcript and detected languages as JSON. If you build an SRT pipeline on OpenAI today, plan to build SRT yourself from timed output before that date. Files can be up to 25 MB.

Deepgram. The pre-recorded response has a words array with a start and end time for every word. Its captions guide builds WebVTT and SRT with the @deepgram/captions (JavaScript) and deepgram-captions (Python) libraries from that result. Multilingual Nova-3 costs more: $0.0052 / min pre-recorded and $0.0092 / min streaming.

AssemblyAI. After a transcript finishes, GET /v2/transcript/{id}/srt or /vtt returns the file, and chars_per_caption sets the maximum length of each caption. Streaming is billed from when the WebSocket opens until it closes, so an idle connection still costs money.

Google Cloud Speech-to-Text. V2 batchRecognize takes an outputFormatConfig with srt and vtt, and writes the files to Cloud Storage. Word times come from enableWordTimeOffsets. The V1 table gives the first 60 minutes a month free, then $0.016 a minute with data logging or $0.024 without. New Google Cloud accounts also get the Free Program's $300 in credits.

Amazon Transcribe. Add a Subtitles block with vtt and srt to a batch job and the files land in your S3 output location. Watch one detail: AWS uses a start index of 0 for subtitle numbering, while most players expect 1. Set OutputStartIndex to 1. Streaming takes FLAC, Ogg Opus or PCM, not MP4.

Rev AI. GET /jobs/{id}/captions returns SRT by default or VTT through the Accept header. Billing rounds up to the nearest second with a 15-second minimum. Whisper Large costs $0.005 per minute, English only.

Live captioning APIs are a different job

A "live captioning API" is a streaming speech-to-text API: you open a WebSocket, send audio as it is spoken, and get text back in pieces, first as guesses and then as final results. OpenAI (gpt-live-transcribe), Deepgram, AssemblyAI, Google, Amazon and Rev AI all document streaming.

The streaming docs we read (OpenAI, Google, Rev AI) describe text coming back in pieces, and Amazon describes subtitles as a batch-job feature. None of them describes an SRT or VTT file coming back from a stream. You get text events with timings, and your player or overlay draws them. If you need a subtitle file of the event afterwards, record it and run the recording through a batch API. For live captions built into Windows, iPhone, Android and Chrome, without any code, see live captions on Windows, iPhone, Android and Chrome.

APIs that fetch existing subtitles

OpenSubtitles API

For films and TV episodes, OpenSubtitles is the usual source. Its help center explains the "opensubtitles api key" question: create an account, then open the API Consumers section of your profile and create a key. Searching has no limit, but downloads do:

  • Anonymous requests: 5 downloads per 24 hours per IP.
  • Signed-in users: from 10 a day for regular users to 1,000 for VIP users, depending on rank.
  • Put your application name and version in the User-Agent header.

More downloads need a paid "pro package"; the help page gives no prices.

YouTube Data API captions

"YouTube subtitle API" usually means the captions resource of the YouTube Data API v3, and it is narrower than people expect. captions.list costs 50 quota units and returns only metadata; the docs say the response "does not contain the actual captions". captions.download costs 200 units, needs OAuth, and "requires the user to have permission to edit the video". It returns SRT, VTT, SBV, SCC or TTML.

So it works for your own channel, not for any public video. With the default 10,000 units a day, that is about 40 videos a day if you list and download each one.

How to choose

  • You want an SRT file with the fewest steps: AssemblyAI or Rev AI (one GET after the job), or Amazon Transcribe if you already work in AWS (set the start index to 1).
  • You want word-level timing to build your own captions: any of the six except Rev AI, where we could not confirm it; on OpenAI only through deprecated whisper-1.
  • You want live captions: a streaming API, and your own rendering.
  • "Subtitle API free": no generation API here is free forever, but free credit goes a long way: Deepgram's $200 credit, AssemblyAI's $50, Rev AI's 5 hours, and 60 free minutes a month at Amazon (first 12 months) and Google (V1).
  • You want subtitles for a film or your own YouTube videos: OpenSubtitles or the YouTube Data API, not speech-to-text.

Before you pick, check your files open where they need to. SRT vs VTT vs TXT explains the formats, and the SRT to VTT converter and SRT editor fix files by hand.

Where CapsAI fits

CapsAI has no public REST API. There is no key to sign up for and no endpoint to call from your backend.

What it has is the CapsAI MCP server, for AI agents such as Claude. An agent connected to it can caption a video file of up to 10 minutes in one of 11 spoken languages, restyle the captions, and call get_transcript to get the video's transcript as SRT subtitles. That is the only way to get an SRT file from CapsAI; the web editor exports one MP4 with the captions burned in. Captioning a video from Claude Code walks through it. CapsAI does not do live captions.

If you only need captions on a few videos, an API is more than you need. Upload the file to video to captions, check the words in the browser and export the MP4. A new account has 3 free minutes, and packs are paid once: $4, $9 or $19 (€4, €9 or €19 in the euro area, ₹49, ₹99 or ₹299 in India).

Published 4 Nov 2026

Frequently asked questions

Which subtitle API should I use?

There is no single answer. To get an SRT or VTT file straight from the API, AssemblyAI, Rev AI, Amazon Transcribe and Google Cloud Speech-to-Text V2 batch all return one. Deepgram gives word timestamps and builds SRT with a helper library. To fetch existing subtitles, use OpenSubtitles for films and TV or the YouTube Data API for your own videos.

Is there a free subtitle API?

Not free forever, but the free credit is generous. As read on 4 November 2026: Deepgram offers a $200 credit, AssemblyAI $50 in credits, Rev AI credits worth 5 hours of Reverb, Amazon Transcribe 60 minutes a month for the first 12 months, and Google's V1 pricing table makes the first 60 minutes a month free.

Does the OpenAI API output SRT or VTT?

Its API reference lists srt and vtt as response formats, but gpt-4o-transcribe and gpt-4o-mini-transcribe return json only, and the guide's subtitle examples use whisper-1. OpenAI says whisper-1 shuts down on 26 February 2027 and recommends gpt-transcribe, which returns JSON.

Is there a live captioning API?

Yes: streaming speech-to-text. OpenAI (gpt-live-transcribe), Deepgram, AssemblyAI, Google, Amazon Transcribe and Rev AI document streaming over a live connection. They return text as it is spoken, which your player draws. To get a subtitle file of a live event, record it and run the recording through a batch API.

How do I get an OpenSubtitles API key?

Create an account on OpenSubtitles, open the API Consumers section of your profile and create a key. Its help center says anonymous requests get 5 downloads per 24 hours per IP and signed-in users from 10 to 1,000 a day by rank, and asks you to put your app name and version in the User-Agent header.

Can the YouTube API download subtitles from any video?

No. captions.download requires OAuth and permission to edit the video, so it works for videos on channels you manage. It costs 200 quota units per call, out of a default 10,000 units a day, and returns SRT, VTT, SBV, SCC or TTML.

Does CapsAI have a subtitle API?

Not a public REST API. People captioning a few videos can upload in the browser and export an MP4 with captions burned in. CapsAI does not do live captions.

See Your Subtitles in Action

Try a style and language for your next video. No sign-up needed to try. Free for a video up to 3 minutes.

Vertical video with sample captions
Upload Your Video

Related CapsAI tools