Skip to main content

audio model comparison

GPT-4o Transcribe vs Scribe v2

GPT-4o Transcribe and Scribe v2 are both current audio models, from OpenAI and ElevenLabs. Scribe v2 is 39% cheaper per audio minute than GPT-4o Transcribe, $0.003667 vs $0.006. On Artificial Analysis AA-WER v2, GPT-4o Transcribe scores 3.96 % WER and Scribe v2 2.18 % WER (Artificial Analysis, measured October 13, 2026).

Data verified October 13, 2026 · Dataset 2026-10-13

Share this

Side by side

Standard-tier prices in effect today. The better value in each row is highlighted.

FactGPT-4o TranscribeScribe v2
Price per audio minute$0.006$0.003667 (better)
Artificial Analysis AA-WER v23.96 % WER (#6 of 6)2.18 % WER (#2 of 6) (better)
ProviderOpenAIElevenLabs
Release dateMarch 20, 2025January 9, 2026
AccessAPI onlyAPI only
Context window16,000 tokensNot reported
Max output2,000 tokensNot reported
Inputtext, audioaudio
Outputtexttext

Scores:Artificial Analysis · 2026-10-13

Open both in the interactive comparison

When to choose each

GPT-4o Transcribe

Best for

  • Speech transcription

Watch out for

  • OpenAI announced its deprecation on 2026-08-26 and recommends GPT Transcribe, which costs $0.0045 per minute against about $0.006 here
Full GPT-4o Transcribe profile

Scribe v2

Best for

  • Batch transcription and diarization

Watch out for

  • $0.22 per hour of audio; keyterm prompting and transcript editing add $0.05 and $0.07 per hour
Full Scribe v2 profile

GPT-4o Transcribe vs Scribe v2 FAQ

Which is cheaper, GPT-4o Transcribe or Scribe v2?

Scribe v2 is 39% cheaper per audio minute than GPT-4o Transcribe, $0.003667 vs $0.006.

Which scores higher on independent benchmarks, GPT-4o Transcribe or Scribe v2?

Scribe v2 leads on Artificial Analysis AA-WER v2, 2.18 % WER to 3.96 % WER. Scores are independent evaluations, not either vendor's launch figures.

Production evaluation

Choosing between GPT-4o Transcribe and Scribe v2?

Public benchmarks narrow the shortlist. We run both on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.

  • A head-to-head evaluation on your real workload
  • Routing between models by cost and difficulty