---
title: "Phonon-2.5 vs ElevenLabs Scribe v2"
description: "ElevenLabs Scribe v2 beats Phonon-2.5 on English word error on all seven Open ASR sets (4.06 % vs 5.06 %). Phonon-2.5 runs on your hardware with no hourly fee."
canonical: "https://www.fermionresearch.com/models/phonon-2-5/vs-elevenlabs-scribe/"
source: "Fermion Research"
---

Phonon/facts checked 13 November 2026

# Phonon-2.5 vs ElevenLabs Scribe v2

Scribe v2 is the more accurate English transcriber of the two on public test sets. Phonon-2.5 is the one you can download, run offline and pay nothing per hour for.

[Phonon-2.5 downloads and benchmarks](https://www.fermionresearch.com/models/phonon-2-5/)

## The short answer

For the lowest English word error from a hosted API, ElevenLabs Scribe v2 is ahead: 4.06 % on the Open ASR Leaderboard's seven English sets, against 5.06 % for Phonon-2.5 and 4.82 % for Phonon-2.5 English, and lower on every one of the seven sets. Choose Phonon-2.5 when audio cannot go to a third-party cloud, when it must work offline or on a device, or when hourly fees are the problem.

ElevenLabs has since released Scribe v3; the leaderboard snapshot we compare against has the Scribe v2 row.

## Side by side

ElevenLabs facts from elevenlabs.io/pricing/api and the ElevenLabs docs, read 2026-11-13.

|  | Phonon-2.5 | ElevenLabs Scribe v2 |
| --- | --- | --- |
| Model type | Open weights on Hugging Face | Hosted API |
| Size | 156 MB download (Phonon-2.5 English: 158 MB) | Not published |
| Languages | 25 European languages (Phonon-2.5 English: English only) | 90+ |
| English word error, seven Open ASR sets | 5.06 % (Phonon-2.5 English: 4.82 %) | 4.06 % |
| Price per audio hour | No per-hour fee when self-hosted | $0.20 (Scribe v2); $0.22 (Scribe v2 Realtime) |
| Where it runs | Your hardware: CPUs (Linux x86-64 and arm64, Windows x86-64), Apple silicon, NVIDIA GPUs, Android | ElevenLabs cloud |
| Streaming | Yes, live WebSocket from fermion serve | Scribe v2 Realtime |
| Diarization | No | Yes |
| Word timestamps | Yes (--json output) | Yes |
| Other | Verbatim output | Audio event tags (laughter, applause); keyterm prompting; files up to 5 GB |

## English, set by set

Scribe v2 is lower on all seven sets. The smallest gap is AMI meetings (8.19 % against 8.76 % for Phonon-2.5 English); the largest relative gaps are Earnings-22 and VoxPopuli.

| Test set | Phonon-2.5 | Phonon-2.5 English | ElevenLabs Scribe v2 |
| --- | --- | --- | --- |
| LibriSpeech clean | 1.60 | 1.64 | 1.06 |
| LibriSpeech other | 3.41 | 3.56 | 2.33 |
| AMI | 9.12 | 8.76 | 8.19 |
| Earnings-22 | 6.72 | 6.22 | 4.80 |
| GigaSpeech | 8.04 | 7.99 | 7.63 |
| SPGISpeech | 3.31 | 3.31 | 3.01 |
| VoxPopuli | 3.23 | 2.28 | 1.41 |
| Average of the seven | 5.06 | 4.82 | 4.06 |

## Where ElevenLabs Scribe is the better choice

- English accuracy: lower word error on every Open ASR set in the snapshot.
- Breadth: more than 90 languages against Phonon-2.5's 25.
- Richer output: speaker diarization, audio event tags and keyterm prompting (up to 1,000 terms in batch).
- Long files: up to 5 GB and 15 hours per request.

## Where Phonon-2.5 is the better choice

- On-device and offline: a 156 MB model that runs on a laptop CPU, a Mac or an Android phone with no network.
- No hourly fee: at $0.20 per hour, 10,000 hours of Scribe v2 is $2,000 at list price; self-hosted Phonon costs the hardware.
- Data control: nothing is sent to a third party, which can settle a compliance review before it starts.
- A pinned model: open weights do not change under you, so results stay reproducible.

```text
pip install fermion-research
fermion transcribe phonon-2.5 interview.wav --json
fermion serve phonon-2.5   # POST /v1/audio/transcriptions, GET /v1/audio/stream
```

## Questions

## Related

- [Phonon-2.5 vs AssemblyAI](https://www.fermionresearch.com/models/phonon-2-5/vs-assemblyai/)
- [Phonon-2.5 vs Meta Muse Voice Transcribe](https://www.fermionresearch.com/models/phonon-2-5/vs-meta-muse/)
- [On-device speech recognition](https://www.fermionresearch.com/deploy/on-device-speech-recognition/)
- [Speech-to-text API cost](https://www.fermionresearch.com/speech/speech-to-text-api-cost/)

## Sources and method

- Scribe prices: elevenlabs.io/pricing/api, read 2026-11-13.
- Scribe languages, diarization, timestamps, keyterms, file limits and Scribe v3: elevenlabs.io/docs/capabilities/speech-to-text and elevenlabs.io/docs/overview/models, read 2026-11-13.
- Rival English word error: Open ASR Leaderboard, seven English sets, snapshot 2026-09-25 (huggingface.co/spaces/hf-audio/open_asr_leaderboard).
- Phonon-2.5 and Phonon-2.5 English: Fermion Research measurement, Phonon-2.5 release post (2026-10-22), Open ASR Leaderboard scoring code, full test sets of the same seven sets.

Canonical: https://www.fermionresearch.com/models/phonon-2-5/vs-elevenlabs-scribe/

Source: [https://www.fermionresearch.com/models/phonon-2-5/vs-elevenlabs-scribe/](https://www.fermionresearch.com/models/phonon-2-5/vs-elevenlabs-scribe/)
