OpenAI has launched two new speech recognition models, GPT Transcribe and GPT Live Transcribe, reducing file transcription costs by 25%. The models scored 3.31% on an independent accuracy benchmark, placing ninth among tracked systems.
TL;DR
- OpenAI introduces two new transcription models, GPT Transcribe and GPT Live Transcribe, with a 25% price reduction for file transcription.
- The models score 3.31% on the AA-WER accuracy benchmark, ranking ninth among competitors.
- Both models accept context, keyword lists, and language hints to improve accuracy on noisy audio.
What happened
OpenAI released GPT Transcribe and GPT Live Transcribe on July 28, 2026, through its developer platform. The company also published model pages, a rewritten transcription guide, and code samples for all supported endpoints.
GPT Transcribe is designed for completed audio files and batch work, processing recordings at approximately 34 times real time. GPT Live Transcribe provides streaming text over an open connection during live conversations.
Both models accept plain text descriptions of the recording, keyword lists, and language hints to enhance accuracy. The file model costs $0.0045 per minute, while the streaming version costs $0.017 per minute of an open session.
Why it matters
The 25% price reduction for file transcription makes OpenAI's services more competitive in the market. However, the accuracy of the models, while improved, still lags behind industry leaders like Alibaba, ElevenLabs, Google, and Mistral.
Developers can tune the latency and transcript stability of GPT Live Transcribe, making it a versatile tool for various applications. Microsoft has already listed both versions in its Foundry platform, indicating early adoption by major tech players.
The launch highlights OpenAI's continued focus on audio services, following the release of Whisper in 2022 and GPT-4o Transcribe in 2025. Despite the improvements, rivals like Mistral have maintained competitive pricing and accuracy.
Key facts
- GPT Transcribe scored a 3.31% word error rate on the AA-WER index, ranking ninth among tracked systems.
- File transcription costs $0.0045 per minute, a 25% reduction from the previous model.
- Streaming transcription costs $0.017 per minute of an open session.
- Both models accept context, keyword lists, and language hints to improve accuracy on noisy audio.
- GPT Transcribe processes recordings at approximately 34 times real time.
- GPT Live Transcribe allows developers to tune latency and transcript stability.
- Microsoft has listed both models in its Foundry platform.
- OpenAI's previous model, GPT-4o Transcribe, cost $0.006 per minute for file transcription.
Context
OpenAI's new transcription models are part of a broader trend in the AI industry to improve speech recognition technology. Accurate and affordable transcription services are in high demand across various sectors, including podcasts, call centers, newsrooms, and research archives.
The launch of GPT Transcribe and GPT Live Transcribe follows OpenAI's previous releases in the audio space, including Whisper and GPT-4o Transcribe. These models have become popular for their affordability and ease of use.
Despite the improvements, OpenAI faces stiff competition from rivals like Alibaba, ElevenLabs, Google, and Mistral, who have maintained higher accuracy and competitive pricing. The AI industry continues to evolve rapidly, with companies constantly pushing the boundaries of speech recognition technology.
