Whisper Tiny

Very small Whisper checkpoint for quick multilingual transcription tests and low-resource speech pipelines.

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

Fast prototypes, edge-style experiments, and low-memory ASR tests where accuracy tradeoffs are acceptable.

Who should use it

Fast prototypes, edge-style experiments, and low-memory ASR tests where accuracy tradeoffs are acceptable.
Builders who want local or self-hosted testing options.

Common workflows

Very lightweight transcription tests
audio workflows
transcription workflows
speech recognition workflows
local workflows

Deployment and hardware notes

Under 100 MB in fp16 — it runs on single-board computers and in browser-side runtimes where nothing larger fits.

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.
Fast prototypes, edge-style experiments, and low-memory ASR tests where accuracy tradeoffs are acceptable.
Useful for voice-activity and language-detection steps in front of a larger model, where being wrong about words costs nothing.

Limitations

Accuracy is low enough that output usually needs human correction; treat it as a latency and integration test rather than a transcription tool.
Under 100 MB in fp16 — it runs on single-board computers and in browser-side runtimes where nothing larger fits.
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

Useful for voice-activity and language-detection steps in front of a larger model, where being wrong about words costs nothing.

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

Platforms: Windows, macOS, Linux

Transcription spec

Memory~0.08 GB in fp16 (38M parameters) · ~0.04 GB in int8Decoder depth4 layersLanguagesMultilingual (99 languages)Audio window30-second audio windows · 448-token cap per window

The smallest Whisper at 4 encoder and 4 decoder layers — the same decoder depth as large-v3-turbo, but paired with a tiny encoder, which is why the two are nowhere near each other in 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~0.08 GB in fp16 (38M parameters)RuntimeTransformers, faster-whisper, whisper.cppContext30-second audio windows · 448-token cap per windowLast updated2026

Exact model card →