Model Overview
GPT-4o Mini Transcribe is a speech-to-text model based on GPT-4o mini, used to transcribe audio into text. OpenAI says that, compared with the original Whisper models, it improves word error rate, language recognition, and transcription accuracy. (developers.openai.com)
Capabilities and Uses
- Accepts audio and outputs transcribed text; it can be used for recording transcription, subtitle drafts, and text records in speech applications.
- Can be accessed through the OpenAI audio transcription API; the official documentation also lists real-time transcription workflows. (developers.openai.com)
Limitations and Status
- Transcriptions may contain errors; important content should be checked against the original audio. The model does not support image or video input and does not output audio.
- The official model documentation lists a 16,000-token context window and a maximum output of 2,000 tokens. (developers.openai.com)
- This model and the specified
gpt-4o-mini-transcribe-2025-03-20snapshot have been marked as deprecated. OpenAI announced thatgpt-4o-mini-transcribewill be shut down on February 26, 2027, and recommends migrating togpt-transcribeorgpt-live-transcribe. Required output formats and metadata should be checked during migration. (developers.openai.com) - The undated model ID is an alias and is not guaranteed to always point to the March 2025 snapshot; the specified snapshot ID can be used to lock the version. (developers.openai.com)
Release Date
The official announcement was published on March 20, 2025; the version snapshot specified in the input is gpt-4o-mini-transcribe-2025-03-20. (openai.com)
API Pricing
According to OpenAI's official transcription model pricing table, prices are in US dollars (USD): $1.25 per 1 million input tokens and $5.00 per 1 million output tokens; the official estimated cost is approximately $0.003/minute, which is an estimate rather than an additional fixed per-minute fee. The pricing table lists prices by model name and does not list a separate price for the 2025 snapshot. Date checked: November 19, 2026. (platform.openai.com)