Whisper large v3
converts spoken audio into text. Sometimes called "ASR" (automatic speech recognition). Deepgram, AssemblyAI.fact-checked, all required fields present, publicly visible
speech.dev has not published an operational assessment for this entry yet. Facts below are source-linked where available.
Overview
- Publisher
- OpenAI
- Lab
- openai
- Last updated
- 2026-05-23
- Modality
- converts spoken audio into text. Sometimes called "ASR" (automatic speech recognition). Deepgram, AssemblyAI.
a voice model trained on multiple languages. Quality varies enormously by language — "supports 40 languages" may be great in 3 and mediocre in the other 37.
multilingual
Pricing
- Billing model
- Per minute
- Rate
- $0.006 per minute
Last verified May 23, 2026
Benchmarks
English short-form WER
Open ASR LeaderboardListed as: openai/whisper-large-v3
- Rank
- #37
- Avg WER
- 7.44%
- RTFx
- 146
Verified 2026-05-23
English long-form WER
Open ASR LeaderboardListed as: openai/whisper-large-v3
- Avg WER
- 11.2275%
Verified 2026-05-23
Multilingual WER
Open ASR LeaderboardListed as: openai/whisper-large-v3
- Avg WER
- 4.81%
Verified 2026-05-23
Technical
- Streaming
- Batch / REST only
- Hosting
- saas-vendor-cloud
- Self-hostable
- Yes
API access
WSSHTTPS
API endpoints
https
SDK languages
python, typescript
Regions (vendor buckets)
Macro areas as published by the vendor (often broad; not a country list).
Compliance (vendor-published)
GDPR DPAHIPAA BAASOC 2 Type II
Suggest an update
Facts-layer corrections only — source URLs required. Opens a GitHub issue; a maintainer runs the content agent after triage. Not for operational notes.
Lab metadata: Update OpenAI · All request types · Full guide