faster-whisper-large-v3

Name: faster-whisper-large-v3
Author: Systran

by Systran

Whisper large-v3 converted to CTranslate2, for transcribing archives rather than clips

Whisper large-v3 weights (converted)params

HuggingFace Run it on your own data, free

Identifiers

Model ID

Systran/faster-whisper-large-v3

Feature URI

Overview

This is not a different model from Whisper large-v3. It is the same weights converted to the CTranslate2 runtime, and the reason it matters is throughput per dollar rather than accuracy.

Transcription is usually the step that decides whether you process an entire back catalogue or only the files someone asks about. Reference Whisper is accurate and expensive enough that teams sample. A faster runtime over identical weights changes that arithmetic without changing what the output says, which is the rare case where you can take the speed and lose nothing.

MIT licensed, multilingual, and it is what most self-hosted transcription pipelines actually run in production.

Architecture

OpenAI Whisper large-v3 weights converted to the CTranslate2 inference engine. The HuggingFace record carries no parameter count because the repository holds a converted format rather than safetensors; the underlying model is Whisper large-v3.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so faster-whisper-large-v3 runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection_id: "col_your_collection",
      documents: [
        {
          document_id: "asset-00412",
          // The model produces text, so it lands in payload. Give the
          // collection a text vector index and embed that text to make it
          // searchable rather than only filterable.
          payload: { extracted_text: modelOutput, source_key: "archive/2026/asset-00412" },
          vectors: { "text-embedding": embeddingOfModelOutput },
        },
      ],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// universal_extractor@v1 runs google/gemini-embedding-2
// (3072-d) over a bucket, with no inference of your own.

Capabilities

Transcription with Whisper large-v3 accuracy
Substantially higher throughput than the reference implementation on the same hardware
Multilingual, with the same language coverage as the source model
Word-level timestamps for locating a moment rather than a file