OpenAI is removing 18 model snapshots effective today, July 23, 2026, marking the first in a series of deprecations that will reshape how developers build AI applications. The company announced these shutdowns on April 22, 2026, giving developers roughly three months to migrate their code to newer alternatives. However, the full scope of changes extends far beyond today, with critical waves scheduled for August, September, October, and December that will affect foundational models many teams still rely on.

What Models Are Disappearing Today?

The 15 models being retired today include specialized variants designed for computer use, web search, text-to-speech, and audio processing. OpenAI is recommending specific replacements for each deprecated model, though not all substitutes perform identically to their predecessors. For example, developers using the search-enabled chat completion models are being directed to GPT-5.4-mini, even though some community members argue that GPT-5-search-api would be a more appropriate replacement.

The audio and text-to-speech migrations have generated particular concern among developers. Multiple forum discussions in 2026 report that the recommended replacement for the March 2025 TTS snapshot, gpt-4o-mini-tts-2025-12-15, truncates final sentences and no longer honors the instructions parameter. This quality regression has prompted some developers to push back against the migration timeline, though OpenAI has not extended the deprecation date.

When Will Larger Waves Hit?

Today's shutdown is just the beginning. OpenAI has scheduled four additional deprecation waves that will remove far more critical infrastructure:

• August 10, 2026: Two chat-latest snapshots (gpt-5.2-chat-latest and gpt-5.3-chat-latest) will be removed, with migration to gpt-5.5.
• August 26, 2026: The Assistants API will be discontinued entirely, replaced by the Responses API and Conversations API.
• September 24, 2026: The Videos API and all Sora model variants will shut down with no replacement listed.
• October 23, 2026: The largest wave removes the entire gpt-3.5-turbo and gpt-4 lines, the original gpt-4o snapshot, and fine-tuned versions built on these models.
• December 11, 2026: Original GPT-5 snapshots and o3 variants will be retired.

Why the October Wave Will Hurt Most Teams?

The October 23 deprecation represents the most disruptive change. Teams that have invested in fine-tuning models like gpt-3.5-turbo, gpt-4, or gpt-4.1-nano face a retraining project rather than a simple code update. Fine-tuned versions of these base models will stop working on the same day, forcing developers to rebuild their custom models from scratch on newer architectures. This is not a sprint; it is a quarter of engineering work for teams with significant fine-tuning investments.

The deprecation also removes gpt-image-1, the image generation model that powered many applications before newer versions arrived. Developers using this model for production image generation will need to migrate to gpt-image-2 by the October deadline.

How to Prepare for OpenAI Model Deprecations

Developers should take immediate action to assess their exposure and plan migrations:

• Audit Your Codebase: Search your repositories for hardcoded model names and snapshot dates, particularly the 2024-12-17 audio snapshots that are being removed today. Identify which deprecated models your applications currently use and prioritize those appearing in the October wave.
• Test Replacement Models: Before migrating to OpenAI's recommended substitutes, test them thoroughly in your environment. For text-to-speech applications, specifically test sentence-final audio output, as the replacement model has known issues with truncation.
• Plan Fine-Tuning Migrations: If your team has invested in fine-tuned models on gpt-3.5-turbo, gpt-4, or other October-wave base models, begin planning retraining projects immediately. Allocate sufficient engineering resources and budget for this work, as it cannot be completed in a single sprint.
• Monitor Official Channels: OpenAI publishes deprecation notices via email to affected customers before updating the public deprecations page. Check your organization's email for notifications and cross-reference them against the official deprecations page, as discrepancies have occurred in the past.

Are OpenAI's Deprecation Notices Reliable?

OpenAI's deprecation communication has shown inconsistencies that developers should be aware of. The April 22 announcement email listed 18 model snapshots, but the public deprecations page currently shows only 15 rows for today's date. The three missing models are all audio variants: gpt-4o-audio-preview-2024-12-17, gpt-4o-mini-audio-preview-2024-12-17, and gpt-4o-mini-realtime-preview-2024-12-17.

Additionally, coverage circulating in July 2026 referenced a July 20 announcement about legacy audio, realtime, and transcription model families being removed on January 20, 2027. However, when the deprecations page was checked on July 23, this announcement did not appear anywhere on OpenAI's official page. The only 2027 date listed concerns fine-tuning job creation on January 6, 2027, not audio deprecation.

This gap between email notifications, public announcements, and the official deprecations page underscores a critical lesson: developers should verify deprecation information through multiple channels rather than relying on secondary summaries or a single source of truth. OpenAI's guarantee of at least six months' notice for generally available models and three months for specialized variants provides some protection, but the inconsistencies in how information is communicated create real risk for teams that do not actively monitor official channels.

What Does This Mean for the Broader AI Development Landscape?

These deprecations reflect OpenAI's rapid model development cycle and its strategy to consolidate its product line around newer, more capable versions. The removal of gpt-3.5-turbo and gpt-4, which powered countless applications since their release, signals that OpenAI views these models as obsolete compared to its newer offerings. However, the lack of a replacement for Sora and the Videos API suggests that OpenAI may be deprioritizing video generation as a product category, at least in its current form.

For developers, the lesson is clear: pinning to specific model snapshots rather than using aliases provides stability, but it also requires active maintenance as OpenAI retires older versions. Teams should establish processes to track deprecation timelines, test migration paths well in advance, and budget for the engineering work required to move off legacy models, particularly those with significant fine-tuning investments.