# ElevenLabs Realtime: Scribe v2 Pricing and Features By Su-Jin Bae · 2026-10-24 · Source: https://www.activepieces.com/blog/elevenlabs-realtime-scribe-v2-pricing-and-features --- ElevenLabs released Realtime, a new speech-to-text API that transcribes audio while the person is still speaking instead of after the recording ends. The failure it targets is familiar: a workflow, whether it's hand-coded or built in a tool like [Activepieces](https://www.activepieces.com), waits for the speaker to finish, so a request sits unprocessed while the person keeps talking. ![Unopened envelopes stacked at a closed door while someone talks beyond it](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/99ecb066-dcd8-44bd-a308-6cfef022398f/elevenlabs-realtime-scribe-v2-pricing-and-featur-0154c450.webp) ElevenLabs Realtime is a speech-to-text API from ElevenLabs that **returns a transcript as audio streams in**, so a workflow can receive and act on spoken words while the person is still talking rather than after the recording ends. ## ElevenLabs Realtime launched: what was released and when Everything here comes from the maker's pages, read on 2026-10-24. When audio goes in continuously and text comes back during speech, that is this streaming speech-to-text capability at work. For automation, a workflow can start on partial input instead of idling until a file arrives, so the delay between speech and action shrinks to the transcription step. This article makes no claims about accuracy, language coverage or latency, because the maker's pages give no figures for any of them. To get access, go through the ElevenLabs API, and treat the maker's API reference as the authority on authentication and request format. Plan requirements and pricing aren't stated here, so confirm them there before building anything that depends on them. ## What realtime speech-to-text is for automation teams If a workflow waits for a recording to end, it can't act on anything said before the end. **Realtime transcription removes that wait**: text reaches the workflow while the person is still speaking. ![Activepieces workflow builder showing a Page Audit step using Text AI with OpenAI GPT-4o to create an SEO audit.](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/06a8a527-00bb-443a-8a42-78ad1fd5fa1a/enterprise-ai-security-framework-for-automation-032ed84e.webp) Say a caller states an account problem in the first sentence and hangs up. With batch transcription, the ticket appears only after the upload and processing finish. The delay is built into the design, because nothing downstream can see the words until the recording is closed. As the maker's pages describe it, ElevenLabs Realtime handles audio as it arrives and returns transcript fragments in sequence, rather than one transcript for the whole file. For a team wiring up automations, that means a trigger can fire on the fragment that contains the relevant phrase, so a call, a voice note or a meeting can start a ticket, a lookup or a notification before the speaker finishes. ![Speaker's words ribbon feeding a machine already outputting a ticket](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/2a912ea5-1c0f-4432-8f42-b15f810b1f7c/elevenlabs-realtime-scribe-v2-pricing-and-featur-205e6454.webp) In the illustration below, two stacked timelines show the difference. The top one is the batch pattern: the recording runs to its end, and only then do a single transcript and its trigger appear. The bottom one is the streaming pattern: audio chunks each yield transcript fragments, and a workflow trigger fires partway through the bar. Take from it that the trigger point moves from after the recording to inside it. ## What is measurably new, from ElevenLabs' own pages Half the information needed for a cost-benefit analysis is missing: the pages state **no latency or accuracy figures**. What they do document is a realtime speech-to-text service reached over WebSocket. Everything below comes from pages checked on 2026-10-24. Read the last row of the table as a gap. It lists what the API reference and pricing page say, and any latency or accuracy claim in this article's workflow advice has to come from your own testing. | Feature | What the page says | Dated source | |---|---|---| | Realtime STT service | Listed as a speech-to-text capability | ElevenLabs API reference, 2026-10-24 | | WebSocket access | Documented | ElevenLabs API reference, 2026-10-24 | | Endpoint | Dedicated realtime endpoint in the reference | ElevenLabs API reference, 2026-10-24 | | Streaming input | Audio is sent while the person speaks | ElevenLabs API reference, 2026-10-24 | | Scribe v2 Realtime | — | ElevenLabs pricing page, 2026-10-24 | | Latency and accuracy | Not stated | ElevenLabs pages, 2026-10-24 | _Prices and plan limits checked against [elevenlabs.io](https://elevenlabs.io/pricing/api) and [elevenlabs.io](https://elevenlabs.io/docs/de/api-reference/speech-to-text/v-1-speech-to-text-realtime) and [github.com](https://github.com/activepieces/activepieces/pull/15981) on October 24, 2026._ Nothing here lets you size a timing budget, so measure first-fragment delay yourself. ### ElevenLabs Realtime concurrency limits by plan **With realtime STT at 6, 30, 45 and 45 concurrent requests across four plans, per the [ElevenLabs Help Center](https://help.elevenlabs.io/hc/en-us/articles/33472885290129), any deployment exceeding 45 concurrent connections can't use this service.** On the lowest plan, the service refuses a seventh simultaneous speaker. The top two plans share a ceiling, so moving up between them adds no realtime capacity. Batch speech to text allows 8, 40, 60 and 60. For TTS Flash/Turbo the limits are 4, 20, 30 and 30, and other TTS models allow 2, 10, 15 and 15, so a voice reply can bottleneck before transcription does. ## What Activepieces has already shipped for ElevenLabs [Activepieces](https://www.activepieces.com/pieces/elevenlabs) has an ElevenLabs integration, and like every connected integration it is available to an AI as a tool. Talk time is the cost driver, and what you pay for stays unchanged whichever way the audio reaches the workflow. ### The dated release and link Activepieces has an ElevenLabs integration, and like every connected integration it is available to an AI as a tool. Being available as a tool means an agent can use the integration directly, without a separate build. That integration is not a separate agent catalog you have to rebuild. Once the ElevenLabs integration is connected in Activepieces, an agent can call it: the same action that runs as a step in a flow is also exposed as a tool on the MCP server each project gets, which Claude, ChatGPT, Cursor or an agent of your own can use. Nothing is exported or wired a second time, and you can read the mechanism in packages/integrations in the MIT-licensed repo. Check that your key's scopes match the actions you enable, because an earlier change, PR 13216 (2026-05-13), updated the API key permissions guidance. The integration itself lists 2 actions and 0 triggers. Nothing in it fires when a person starts speaking, and it has no step named for Realtime. ### Connecting it in a workflow today You can reach Realtime through an HTTP step. Because the integration has no trigger, something else must start the flow, such as a webhook or a schedule. The HTTP step then calls ElevenLabs directly, authenticated with the same API key. ![A workflow builder showing a schedule trigger configured to run every Friday at 3 pm UTC.](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/6cfe18ef-b67b-43f6-b8bc-272af9e74826/elevenlabs-realtime-scribe-v2-pricing-and-featur-b82668c5.webp) We didn't run or test Realtime through this path. This describes what the pages show, not a verified build. The consequence is that you own the connection handling the integration would otherwise hide. ## What Realtime's pricing and speed offer An open connection. It buys transcripts delivered while audio is still arriving, not a measured gain in speed or accuracy. ### ElevenLabs Realtime speech-to-text pricing per hour ElevenLabs Realtime delivers streaming text while the person speaks, and no independent measurement has been run, leaving no external validation of the streaming performance, so users cannot independently verify that real-time transcription works as claimed. The batch flow returns text after the recording ends; its price isn't restated here, and it wasn't measured either. Budget by audio hours, then, not by the number of workflow runs they trigger. The batch price is left out because this comparison rests only on the Realtime listing. ### Streaming versus batch transcription ![Ten audio reels beside one invoice slip and a few run tokens](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/93c85f66-dae1-48b4-83aa-4e3810bb2ef3/elevenlabs-realtime-scribe-v2-pricing-and-featur-c7b56cdd.webp) According to the [ElevenLabs API reference](https://elevenlabs.io/docs/de/api-reference/speech-to-text/v-1-speech-to-text-realtime), Realtime is a WebSocket API for streaming audio input and receiving transcription results. With a batch flow, **text returns only once the recording is complete**, so a downstream step can't start until the speaker has stopped. Streaming removes that wait, but no latency or accuracy test backs any speed claim here. Any claim that Realtime is faster or more accurate in your workflow is untested, and a pilot on your own audio would have to settle it. ## What to watch next with ElevenLabs Realtime On the maker's own pages, five things need checking, each logged with a date. A pilot on your audio is only meaningful if you know which rate and options it ran against. ### What ElevenLabs Realtime is Words can reach a workflow while the person is still talking, because this is ElevenLabs' speech-to-text capability for live audio. The pricing page lists it as Scribe v2 Realtime. Pages change without notice, so a pilot priced on one day can quietly diverge from what you're billed on another. To turn that risk into a routine, run the checkpoints below in order and keep the results with the pilot notes, so a surprise on an invoice or a rejected parameter can be traced to a change instead of guessed at: 1. Open the [ElevenLabs pricing page](https://elevenlabs.io/pricing/api) and note the Scribe v2 Realtime rate. 2. Compare the API reference endpoint and its parameters against what your integration sends. 3. Read the docs changelog for entries since your last check. 4. Record the supported options, including any concurrency limits the docs state. 5. Log the check date next to everything above. None of this predicts what ElevenLabs will change. It only fixes what was true on the day you looked. On the day you check, the route to access is whatever the API reference and pricing page describe, and this article doesn't restate it. Treat checkpoint two as the authority: if the documented endpoint or parameters differ from your last record, fix the integration before you run the pilot. ## Frequently asked questions Whether ElevenLabs Realtime is free to try isn't settled by the maker's pages, as checked on 2026-10-24, so treat it as unconfirmed. The pricing unit is also unclear, so you can't yet estimate what a pilot would cost. Confirm both on the maker's pricing page before you commit budget. ElevenLabs made it. It's a speech-to-text capability from that company, and the maker's own pages are the only source this article relies on for what it does. If you want transcripts in other apps, you can do that, but only through an integration you build against the documented API. I couldn't confirm a ready-made connector or trigger on the maker's pages. Whether access runs over a WebSocket is something to verify against the documented endpoint, not assume. The API reference also gives no latency figure, so measure the delay inside your own workflow rather than trusting a number you can't source. For the official documentation, start from the ElevenLabs site and its speech-to-text material. I'm not giving a deep link, because I didn't verify one. Use that documentation as the authority for endpoints and parameters, and re-check it before each pilot. ## Related reading - [Get Realtime Notifications About Stripe Failed Payments on Slack](https://www.activepieces.com/blog/get-realtime-notifications-about-stripe-failed-payments-on-slack) - [AI Launch Week Day 2: Universal Text AI](https://www.activepieces.com/blog/text-ai) - [Best ElevenLabs Alternatives in 2026 Compared](https://www.activepieces.com/blog/best-elevenlabs-alternatives-in-2026-compared) ## References - [ElevenLabs Help Center](https://help.elevenlabs.io/hc/en-us/articles/33472885290129)