GPT-4o Mini Transcribe (Azure)

Microsoft announced the availability of OpenAI's GPT-4o audio models — GPT-4o-Transcribe, GPT-4o-Mini-Transcribe, and GPT-4o-Mini-TTS — in Microsoft Foundry Models, with a practical developer guide for wiring them into applications. GPT-4o-Mini-Transcribe is positioned as the fastest of the two speech-to-text models with "great" quality and streaming support. The post lists targeted audio pretraining, advanced distillation from larger models, and reinforcement learning as the technical advances behind transcription accuracy and reduced misrecognition. It recommends rapid transcription for live captioning, quick-response apps, and budget-sensitive scenarios, then walks through steps: obtaining an Azure OpenAI endpoint and key, authenticating with Azure CLI, storing credentials in an .env file, installing dependencies, and testing with a Gradio deployment.

Microsoft introduced the public preview of Azure OpenAI's advanced audio models, including GPT-4o-Mini-Transcribe, deployable immediately in the East US2 region on Azure AI Foundry. In the comparison table, the mini transcribe model is rated for great quality and the fastest speed, accepting text and audio input, returning text output, and supporting streaming. The announcement attributes the gains to targeted audio pretraining on specialized datasets, distillation that transfers knowledge from larger models into efficient ones, and reinforcement learning that improves transcription accuracy and reduces misrecognition in complex recognition tasks. Suggested use cases for GPT-4o-Mini-Transcribe include live captioning, quick-response applications, and cost-sensitive workloads, with a TTS demo repository provided for experimentation.

In a Microsoft Learn Q&A thread, an Azure OpenAI moderator stated on March 21, 2025 that gpt-4o-transcribe and gpt-4o-mini-transcribe were not available yet in Azure OpenAI, even though the models were compatible with the Realtime API. The reply advised monitoring the What's New section for future additions. The thread captures a dated Azure-side availability snapshot for the exact subject variant shortly after OpenAI's launch announcement, with Microsoft saying it was continuously working to bring new OpenAI models to the service. The remaining replies are moderator follow-ups rather than substantive technical updates.

OpenAI announced a new suite of audio models in the API on March 20, 2025, including the speech-to-text models gpt-4o-transcribe and gpt-4o-mini-transcribe, making them available to developers worldwide. The post was updated on August 28, 2025, to note general availability of the Realtime API. OpenAI says the latest speech-to-text models set a new state-of-the-art benchmark for accuracy and reliability, particularly with accents, noisy environments, and varying speech speeds. The same release adds instruction-following text-to-speech, letting developers specify a voice style such as a sympathetic customer service agent.

Microsoft Foundry's model catalog lists gpt-4o-mini-transcribe as a generally available, OpenAI-published speech-to-text model, currently at version 2025-12-15 and offered as a Direct from Azure model under a single Azure license with unified billing and PTU portability. The listing describes it as a cost-efficient transcription model built for accurate, fast audio-to-text conversion. The catalog entry records the model's technical envelope for developers: text and audio input, text output, a 16,000-token context window, and a 4,096-token output limit. It also highlights improved word error rate and language recognition, particularly for accents, noisy environments, and varying speech speeds, making the model a fit for quick, reliable transcription pipelines.