Whisper Medium

Mid-sized Whisper checkpoint for multilingual transcription and translation with a more practical hardware footprint than the large variants.

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

Builders balancing local transcription quality and runtime cost for meetings, media, and batch audio processing.

Who should use it

Builders balancing local transcription quality and runtime cost for meetings, media, and batch audio processing.
Builders who want local or self-hosted testing options.

Common workflows

Balanced transcription workflows
audio workflows
transcription workflows
speech recognition workflows
local workflows

Deployment and hardware notes

~1.5 GB in fp16 runs on almost any GPU, and int8 makes CPU-only transcription practical for short files.

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.
Builders balancing local transcription quality and runtime cost for meetings, media, and batch audio processing.
Worth benchmarking against large-v3-turbo before committing: on similar hardware the turbo model usually wins on both speed and accuracy.

Limitations

Largely displaced by large-v3-turbo, which occupies the same memory class with a far shallower decoder. Accuracy sits clearly below the large checkpoints on difficult audio.
~1.5 GB in fp16 runs on almost any GPU, and int8 makes CPU-only transcription practical for short files.
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

Worth benchmarking against large-v3-turbo before committing: on similar hardware the turbo model usually wins on both speed and accuracy.

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

Platforms: Windows, macOS, Linux

Transcription spec

Memory~1.5 GB in fp16 (764M parameters) · ~0.8 GB in int8Decoder depth24 layersLanguagesMultilingual (99 languages)Audio window30-second audio windows · 448-token cap per window

A genuinely smaller model rather than a trimmed large: 24 encoder and 24 decoder layers at a narrower width. Similar memory to large-v3-turbo, but slower to decode with its full-depth decoder.

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~1.5 GB in fp16 (764M parameters)RuntimeTransformers, faster-whisper, whisper.cppContext30-second audio windows · 448-token cap per windowLast updated2026

Exact model card →