Together AI Launches Speech-to-Text: High-Performance Whisper APIs

Together AI Launches Speech-to-Text: High-Performance Whisper APIs

⚡️ FlashAttention-4: up to 1.3× faster than cuDNN on NVIDIA Blackwell →

Introducing Together AI's new look →

🔎 ATLAS: runtime-learning accelerators delivering up to 4x faster LLM inference →

⚡ Together GPU Clusters: self-service NVIDIA GPUs, now generally available →

📦 Batch Inference API: Process billions of tokens at 50% lower cost for most models →

🪛 Fine-Tuning Platform Upgrades: Larger Models, Longer Contexts →

All blog posts

Model Library

Published 7/10/2025

Together AI Launches Speech-to-Text: High-Performance Whisper APIs

Authors

Rajas Bansal, Rishabh Bhargava, Sonny Khan

Table of contents

40+ Models Chosen for Production...40+ Models Chosen for Production...40+ Models Chosen for Production...

Links in this article

LINKS IN THISARTICLE Whisper large-v3 Speech-to-text docs ‍ Transcription API Docs ‍ Translation API Docs Try now in Playground Contact us for Enterprise ‍ Get notified

‍

Together AI's First Voice Offering: High-Performance Transcription at Scale Today marks an important expansion for Together AI. We're launching our speech-to-text APIs that solve the fundamental problem holding back voice applications: speed of high-quality transcription and translation. Most developers building voice features hit the same wall. Existing transcription services are simply too slow for real-world applications. For longer audio, they're forced into complex chunking workflows that introduce errors and degrade quality. When audio processing becomes a bottleneck, entire categories of applications become impossible. Performance That Changes What You Can Build Our Whisper V3 Large deployment delivers transcription 15x faster than OpenAI while maintaining full accuracy. This performance comes from several key optimizations: smart voice activity detection using Silero for precise audio segmentation, intelligent chunking and batching strategies for longer audio files, and engine improvements to the Whisper model itself that maximize GPU utilization. This isn't just a technical improvement—it's the difference between transcription as a batch process and transcription as a building block for real-time applications. Consider what becomes possible when transcription happens in seconds rather than minutes. Customer support calls analyzed in real-time Meeting insights delivered before participants leave the room Voice agents that respond naturally instead of asking users to wait Medical scribes that keep pace with doctor-patient conversations

We've also eliminated the practical limitations other services impose. While OpenAI caps uploads at 25MB, we handle files exceeding 1GB. Our infrastructure processes 30+ minute calls seamlessly at $0.0015 per audio minute - delivering substantial cost savings for high-volume applications. Production-Ready API Design Our speech-to-text APIs ship with capabilities designed for real deployment scenarios: Enterprise-scale file handling - process files exceeding 1GB compared to OpenAI's 25MB limit, with support for 30+ minute audio without chunking Superior word-level alignment - advanced model delivers the highest quality timestamps available, outperforming OpenAI Comprehensive language support - transcription and translation across 50+ languages with automatic detection Dedicated endpoints - reserved GPU capacity for sub-second processing speeds beyond our already-fast serverless offering Batch processing - handle large async workloads with consistent performance for high-volume applications

Our interactive playground lets you test transcription quality immediately with your own audio files. No setup required, no complex integration to validate fit. Upload, process, see results in real-time.

Building Toward Complete Voice Infrastructure Voice AI applications in education, customer success, and interactive agents all face the same fundamental challenge: accumulated latency and quality issues across fragmented speech pipelines. When transcription, reasoning, and response generation happen across multiple providers, the delays compound into user experiences that feel sluggish and unnatural. Many Together AI customers already use our LLM APIs for conversational applications, from customer support automation to educational tools. They've been requesting voice capabilities to make these experiences more natural and accessible. Adding high-performance speech-to-text establishes the foundation for voice-enabled applications while eliminating a major bottleneck. Available Now Our speech-to-text APIs are live today through our standard endpoints. Existing customers can add transcription using the same authentication and billing they're familiar with. We've designed for compatibility with existing Whisper integrations, minimizing migration effort. Visit our interactive playground to test with your audio files, review our speech-to-text documentation for integration details, and explore our transcription and translation API references. Experience transcription that actually works at application scale - the future of voice applications isn't limited by transcription speed anymore. Use our Python SDK to quickly integrate Whisper into your applications:

from together import Together

Initialize the client

client = Together()

Basic transcription

response = client.audio.transcriptions.create( file="path/to/audio.mp3", model="openai/whisper-large-v3", language="en" ) print(response.text)

Basic translation

response = client.audio.translations.create( file="path/to/foreign_audio.mp3", model="openai/whisper-large-v3" ) print(response.text)