Models / OpenAI
Transcribe

Whisper Large v3

State-of-the-art automatic speech recognition and translation model supporting 99 languages with 10-20% error reduction over previous versions.

About model

Whisper Large v3 has 1.55B parameters. On Together AI it costs $0.0015 per min.

Whisper Large v3 is OpenAI's open speech recognition model, released in November 2023. It transcribes speech in about 100 languages and can translate speech into English, detecting the language of the audio automatically. The model has 1.55 billion parameters and was trained on 5 million hours of audio: 1 million hours of weakly labeled audio and 4 million hours labeled by Whisper Large v2. Compared with Large v2 it uses a finer audio input (128 Mel frequency bins, up from 80) and adds Cantonese, and OpenAI reports 10% to 20% fewer errors across a wide range of languages. Available on Together AI.

Hours of Training Audio

5M

1 million hours weakly labeled plus 4 million hours pseudo-labeled

Fewer Errors

10-20%

Reduction in errors OpenAI reports compared with Whisper Large v2

Languages

~100

Transcription with automatic language detection

Model key capabilities
  • Multilingual Transcription: Speech to text in about 100 languages, with the language detected automatically
  • Speech Translation: Translates speech from other languages into English text
  • Robust to Real-World Audio: Improved robustness to accents, background noise and technical language
  • Production-Ready Infrastructure: 99.9% SLA, available on Together AI serverless infrastructure
  • API usage

    • cURL
    • Python
    • Typescript

    Endpoint:

    openai/whisper-large-v3

    curl -X POST "https://api.together.ai/v1/audio/transcriptions" \
      -H "Authorization: Bearer $TOGETHER_API_KEY" \
      -F "file=@audio.mp3;type=audio/mpeg" \
      -F "model=openai/whisper-large-v3"
    from pathlib import Path
    
    from together import Together
    
    client = Together()
    
    response = client.audio.transcriptions.create(
        file=Path("audio.mp3"),
        model="openai/whisper-large-v3",
    )
    print(response.text)
    import { createReadStream } from 'fs';
    import Together from 'together-ai';
    
    const together = new Together();
    
    const transcription = await together.audio.transcriptions.create({
      file: createReadStream('audio.mp3'),
      model: 'openai/whisper-large-v3',
    });
    console.log(transcription.text);
  • Model card

    Performance Architecture:
    • Whisper V3 Large deployment delivering transcription 15x faster than OpenAI
    • Smart voice activity detection using Silero for precise audio segmentation
    • Intelligent chunking and batching strategies optimized for longer audio files
    • Advanced GPU utilization optimizations maximizing processing efficiency
    • Sub-second processing speeds with dedicated endpoint infrastructure

    Technical Capabilities:
    • Enterprise-scale file handling supporting files exceeding 1GB vs competitors' 25MB limits
    • Superior word-level alignment delivering highest quality timestamps available
    • Comprehensive language support across 50+ languages with automatic detection
    • Seamless processing of 30+ minute audio without complex chunking workflows
    • Batch processing capabilities for large async workloads with consistent performance

    Infrastructure Design:
    • Production-ready API design built for real deployment scenarios
    • Reserved GPU capacity for guaranteed processing speeds
    • Cost-effective pricing at $0.0015 per audio minute for high-volume applications
    • Compatible with existing Whisper integrations for minimal migration effort
    • Serverless and dedicated endpoint options for different performance requirements

    ‍

  • Applications & use cases

    High-Speed Processing Applications:
    • Customer support call analysis with rapid post-call insights
    • Meeting transcription delivered quickly after recording completion
    • Medical transcription services with efficient workflow processing
    • Content transcription for accessibility and media creation

    Enterprise Solutions:
    • High-volume call center transcription and analysis workflows
    • Educational platforms with voice-enabled learning and assessment tools
    • Compliance and quality assurance audio documentation for regulated industries
    • Large-scale content processing for media and entertainment companies
    • Corporate training and onboarding with automated audio transcription

    Voice-Enabled Applications:
    • Conversational AI systems requiring accurate speech input processing
    • Voice-controlled interfaces for accessibility and hands-free operation
    • Multilingual communication platforms with translation capabilities
    • Content creation tools for podcasts, videos, and audio content
    • Voice analytics and sentiment analysis for customer experience optimization

    Developer Integration:
    • Foundation layer for voice AI application development
    • Building block for voice-enabled customer support automation
    • Integration component for educational technology platforms
    • Core infrastructure for voice assistant applications
    • API component for adding speech capabilities to existing applications

    ‍

  • Released
    November 7, 2023
  • Last updated
    July 9, 2025
  • External link
  • Category
    Transcribe