Whisper large-v3
The default speech-to-text: 100 languages, Apache-2.0, everywhere.
About
Whisper large-v3 (openai/whisper-large-v3) is OpenAI's 1.54B-parameter speech recognition model — the de-facto standard for transcription, translation, and language detection across 100 languages. Released November 2023, it's Apache-2.0 licensed (genuinely open weights), not gated, with 3.7M Hugging Face downloads. The ecosystem is vast: faster-whisper (CTranslate2 speed), insanely-fast-whisper, distil-whisper, and countless UIs. For real-time use, prefer whisper-large-v3-turbo or distilled variants; large-v3 itself remains the accuracy reference. Known quirk: occasional hallucinations on silence/noise.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (OpenAI, 2023): 10-20% relative WER reduction over large-v2 across languages; top of the HF ASR leaderboard on release — provider-reported. Independently re-benchmarked extensively since.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: HF downloads 3,688,803 (peer-normalized); released 1071 days ago. Manual review: default STT (1.5B, 100 langs, 3.7M HF dl), vast ecosystem; slower than turbo/distilled for real-time; silence hallucinations are a known quirk. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.5/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · HF downloads 3,688,803 (peer-normalized)
- · released 1071 days ago
- · manual review: default STT (1.5B, 100 langs, 3.7M HF dl), vast ecosystem; slower than turbo/distilled for real-time; silence hallucinations are a known quirk