Whisper Small

Smaller Whisper checkpoint for multilingual transcription and translation on consumer hardware with lower memory needs than medium or large models.

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

Local transcription setups that need a lighter model for captions, notes, and general speech-to-text tasks.

Who should use it

Local transcription setups that need a lighter model for captions, notes, and general speech-to-text tasks.
Builders who want local or self-hosted testing options.

Common workflows

Local transcription with lighter resource needs
audio workflows
transcription workflows
speech recognition workflows
local workflows

Deployment and hardware notes

Under half a gigabyte in fp16, so it transcribes on CPU at usable speed and leaves a small GPU almost entirely free.

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.
Local transcription setups that need a lighter model for captions, notes, and general speech-to-text tasks.
A good fit for on-device or embedded transcription where the alternative is sending audio off the machine.

Limitations

Error rates climb noticeably on accented speech, technical vocabulary, and noisy recordings; proper nouns are the first thing it gets wrong.
Under half a gigabyte in fp16, so it transcribes on CPU at usable speed and leaves a small GPU almost entirely free.
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

A good fit for on-device or embedded transcription where the alternative is sending audio off the machine.

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

Platforms: Windows, macOS, Linux

Transcription spec

Memory~0.5 GB in fp16 (242M parameters) · ~0.24 GB in int8Decoder depth12 layersLanguagesMultilingual (99 languages)Audio window30-second audio windows · 448-token cap per window

12 encoder and 12 decoder layers — the smallest Whisper most people find usable for real multilingual transcription, and the usual floor for CPU-only pipelines that still need reasonable accuracy.

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~0.5 GB in fp16 (242M parameters)RuntimeTransformers, faster-whisper, whisper.cppContext30-second audio windows · 448-token cap per windowLast updated2026

Exact model card →