# Is Scribe v3 Live SOTA?

Pending benchmarks. Announced Nov 10, 2026; no independent results yet.

As of Nov 22, 2026, Scribe v3 Live is pending benchmarks on isitsota: announced Nov 10, 2026, with no independent results on the boards we track yet.

ElevenLabs' realtime speech-to-text model, announced November 10, 2026. Available in ElevenAPI as scribe_v3_live and in ElevenAgents, and it is a default in the ElevenLabs app alongside Scribe v3. Official product pages state a median latency of about 150 ms. Features include speaker diarization, turn prediction, a multilingual mode, a keyterm library, entity detection, transcript editing and word-level timestamps. No public weight release is stated.

Canonical: https://isitsota.com/m/elevenlabs-scribe-v3-live
Last updated: 2026-11-10
Tracking since: 2026-11-10

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | ElevenLabs |
| Announced | Nov 10, 2026 |
| API | Live (`scribe_v3_live`) |
| Open weights | Closed · No public weight release stated |
| Type | Realtime speech-to-text model for streaming audio; not the batch Scribe v3 model (ElevenLabs blog and docs) |
| Hosted access | ElevenAPI as scribe_v3_live, and in ElevenAgents. Official blog: both Scribe v3 and Scribe v3 Live are the default in the ElevenLabs app. Official docs note it is an option in the agents platform, not yet the default there (ElevenLabs blog and product page) |
| Latency | Official product pages state a median latency of about 150 ms (ElevenLabs realtime Speech to Text page) |
| Languages | Official docs and product pages list 90+ languages. The launch blog reports results across 102 languages on an internal version of FLEURS (ElevenLabs blog and docs) |
| Features | Speaker diarization in realtime, turn prediction from conversation context (the same turn detection ElevenAgents uses), multilingual mode for mid-sentence language switches, keyterm prompting, a keyterm library of up to 10,000 terms, entity detection, transcript editing, and word-level timestamps (ElevenLabs blog and Speech to Text docs) |
| Pricing | Official ElevenAPI price: $0.22 per hour of audio, including silence. Keyterm prompting $0.08 per hour, keyterm library $0.08 per hour, and transcript editing $0.12 per hour on the public pricing table (ElevenLabs API pricing) |
| Medical variant | The launch blog says a version of Scribe v3 Live optimized for medical conversations is also offered in ElevenAgents. No separate public model ID is given on that page (ElevenLabs blog) |
| Deployment | Runs in the US by default; isolated EU, India and Singapore environments for enterprise customers with data residency requirements. Zero Retention Mode, SOC 2, ISO 27001 and PCI DSS Level 1 coverage, and HIPAA-eligible with a Business Associate Agreement, are stated on the launch blog; private deployment is also offered (ElevenLabs blog) |
| License | Not stated; no public weights |
| Related | Separate from Scribe v3 (the batch model), Scribe v2, Scribe v2 Realtime, Cue and Eleven Realtime |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [x] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: self-reported 43% lower semantic WER and 31% lower WER than Scribe v2 Realtime across 102 languages on an internal version of FLEURS, with an improvement in every language tested (ElevenLabs blog). Official Speech to Text docs state 44% lower semantic WER than Scribe v2 Realtime on FLEURS.
- Lab claim: self-reported 3.1% diarization error rate on the English portion of ElevenLabs' 16-language real-world dataset, compared with 5.9% to 7.3% for the other realtime models it tested, after aligning each model's transcript to the reference; the blog says these figures cannot be compared with its batch diarization results (ElevenLabs blog).
- Lab claim: self-reported 39% fewer diarization errors than the next realtime model ElevenLabs tested (ElevenLabs blog).

## Sources

- LAB: [Introducing Scribe v3 and v3 Live (ElevenLabs blog)](https://elevenlabs.io/blog/scribe-v3) (2026-11-10)
- LAB: [ElevenLabs docs: Speech to Text](https://elevenlabs.io/docs/overview/capabilities/speech-to-text)
- LAB: [ElevenLabs realtime Speech to Text product page](https://elevenlabs.io/realtime-speech-to-text)
- LAB: [ElevenAPI pricing](https://elevenlabs.io/pricing/api)
- LAB: [ElevenLabs on X: Scribe v3](https://x.com/ElevenLabs/status/2120217467100541396) (2026-11-10)
- LAB: [ElevenLabs developers on X: Scribe v3](https://x.com/ElevenLabsDevs/status/2120219053046595748) (2026-11-10)

Tracking continues until an independent board lists it.

## Related

- Lab: [ElevenLabs](https://isitsota.com/labs/elevenlabs.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/elevenlabs-scribe-v3-live
