OpenAI sets February 2027 shutdown for Whisper-1 and GPT-4o transcription models

OpenAI has set a firm end date for four of its transcription models, giving developers roughly six months to migrate off Whisper-1 and the GPT-4o transcription family before they stop working in the API.

What's new

Per OpenAI's API deprecations page: "On August 26, 2026, we notified developers using whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize of their deprecation and removal from the API on February 26, 2027."

All four models share the same shutdown date and the same recommended replacements — gpt-live-transcribe or gpt-transcribe — regardless of which of the four a developer is currently using. After February 26, 2027, calls to any of the four will fail outright rather than degrade gracefully.

Context

Whisper-1 is one of OpenAI's oldest and most widely embedded API models — the original open-and-hosted speech-to-text model many transcription products were built around before OpenAI shipped GPT-4o's audio variants. The gpt-4o-transcribe family arrived later as a higher-accuracy, real-time-capable alternative. Retiring all four together, rather than staggering them, suggests OpenAI now considers the newer gpt-live-transcribe/gpt-transcribe line a full replacement for both the legacy and the interim generation.

The move sits alongside other retirements OpenAI has scheduled for the same window: legacy audio, realtime, and gpt-4o-audio variants are set to shut down in January 2027, and GPT-5/o3 snapshots in December 2026 — a broader clearing-out of pre-GPT-6-era model variants across modalities.

Why it matters

Whisper-1 in particular has years of accumulated integrations — transcription pipelines, meeting-notes tools, and voice products that treat it as a stable, low-cost default. A six-month notice window is standard for OpenAI, but the breadth of this deprecation (four models, one cutover date, two replacement targets) means teams running any of the four need to actually test gpt-live-transcribe or gpt-transcribe for accuracy and latency parity well before February, rather than assuming a drop-in swap. Missing the window means transcription calls failing in production with no fallback.