Skip to content
AI Atlas
Family

Whisper

All 13 models in the terminal →

first seen 11 Sept 2026

family_01M2A55TSFT7DS7232RWE6PZ2C

Models
13
First release
26 Sept 2022
Latest release
1 Oct 2024
Parameters
37.8M – 1.54B
Artifacts
2
Providers
0

Releases

Family releases by statusActiveDeprecatedOct 22Jan 23Apr 23Jul 23Oct 23Jan 24Apr 24Jul 24whisper-base — 26 Sept 2022whisper-base.en — 26 Sept 2022whisper-large — 26 Sept 2022whisper-medium — 26 Sept 2022whisper-medium.en — 26 Sept 2022whisper-small — 26 Sept 2022whisper-small.en — 26 Sept 2022whisper-tiny — 26 Sept 2022whisper-tiny.en — 26 Sept 2022whisper-large-v2 — 5 Dec 2022Whisper — 7 Nov 2023Whisper Large V3 Turbo — 1 Oct 2024

12 dated releases · dot size = best benchmark rank of the member · 1 member without a sourced release date not shown

Members 13

Family members
ModelParamsContextOpennessLicenceReleasedStatusBest rankCompare
Whisper Large V3 TurboOpenAI808.9M—Open weightsMIT1 Oct 2024Active—
WhisperOpenAI1.54B—ClosedApache-2.07 Nov 2023Deprecated—
whisper-large-v2OpenAI1.54B—Open weightsApache-2.05 Dec 2022Active—
whisper-baseOpenAI72.6M—Open weightsApache-2.026 Sept 2022Active—
whisper-base.enOpenAI72.6M—Open weightsApache-2.026 Sept 2022Active—
whisper-largeOpenAI1.54B—Open weightsApache-2.026 Sept 2022Active—
whisper-mediumOpenAI763.9M—Open weightsApache-2.026 Sept 2022Active—
whisper-medium.enOpenAI763.9M—Open weightsApache-2.026 Sept 2022Active—
whisper-smallOpenAI241.7M—Open weightsApache-2.026 Sept 2022Active—
whisper-small.enOpenAI241.7M—Open weightsApache-2.026 Sept 2022Active—
whisper-tinyOpenAI37.8M—Open weightsApache-2.026 Sept 2022Active—
whisper-tiny.enOpenAI37.8M—Open weightsApache-2.026 Sept 2022Active—
gpt-realtime-whisperOpenAI——Closed——Active—

Lineage among members 1 relations

Open in Graph →
Lineage of the Whisper familyWhisper Large V3 Turbo fine-tuned from WhisperWhisper1.54BWhisperWhisper Large V3 Turbo808.9MWhisper Large V3 Turbo

Benchmark progress

Benchmark progress needs at least two dated members with a rank on the same benchmark. Ranks (not scores) are what the family endpoint provides.

Aggregates are computed from the members' sourced claims; members without a release date or parameter count are left out of the range and timeline rather than guessed. Methodology →