Whisper Large V2

Large Whisper checkpoint for multilingual speech recognition and speech-to-English translation when transcription accuracy matters more than small-model efficiency.

OpenAI · Whisper

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedAugust 2026SourcesExact model card

Model checkpoints, context windows, provider support, local runtime compatibility, and license terms can change quickly. Verify the exact model card before production or commercial use.

Best for

Teams running accuracy-first transcription, subtitles, meeting notes, or archive workflows on multilingual audio.

Who should use it

Teams running accuracy-first transcription, subtitles, meeting notes, or archive workflows on multilingual audio.
Builders who want local or self-hosted testing options.

Common workflows

Accuracy-focused transcription baseline
audio workflows
transcription workflows
speech recognition workflows
local workflows

Deployment and hardware notes

~3.1 GB in fp16, ~1.5 GB in int8 — the same footprint as large-v3, since the parameter count is effectively unchanged.

License and usage notes

Apache 2.0. Open source weights and code. Verify the exact model card and license terms for the checkpoint or hosted provider you use.

Strengths

Open source weights and code model option for Whisper workflows.
Teams running accuracy-first transcription, subtitles, meeting notes, or archive workflows on multilingual audio.
Swapping to large-v3 is a checkpoint change with no pipeline change, so an A/B on your own audio is cheap. Note that v3 expects 128-bin features, which the processor handles automatically.

Limitations

Carries large-v3's full decoding cost without its accuracy gains, so a new project has little reason to start here. Some languages regressed between v2 and v3, which is the one case for staying.
~3.1 GB in fp16, ~1.5 GB in int8 — the same footprint as large-v3, since the parameter count is effectively unchanged.
Context window and limits: 30-second audio windows · 448-token cap per window.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Swapping to large-v3 is a checkpoint change with no pipeline change, so an A/B on your own audio is cheap. Note that v3 expects 128-bin features, which the processor handles automatically.

Local runtimes: Transformers, faster-whisper, whisper.cpp

Platforms: Windows, macOS, Linux

Transcription spec

Memory~3.1 GB in fp16 (1.5B parameters) · ~1.5 GB in int8Decoder depth32 layersLanguagesMultilingual (99 languages)Audio window30-second audio windows · 448-token cap per window

The previous large generation, identical in size and shape to large-v3 (32 encoder and 32 decoder layers) but trained on 80 mel bins. Superseded for new work; still widely pinned in existing pipelines.

Sources to verify

Exact model card

Related resources

Continue with model source notes, local tools, and implementation guides related to this model.

Whisper family hub AI tools directory Model source notes Models directory Guides library

Hardware~3.1 GB in fp16 (1.5B parameters)RuntimeTransformers, faster-whisper, whisper.cppContext30-second audio windows · 448-token cap per windowLast updated2026

Exact model card →