AI Media Model Retirements Tracker: Shutdown Dates and Replacements by Provider

Image, video, speech and music models are being retired faster than most teams update their integrations, and a retired model id simply stops answering. This tracker collects announced deprecation and shutdown dates, and the replacements providers name, for AI media generation models: image (DALL-E, GPT Image, Imagen, Nano Banana, FLUX, Stable Diffusion), video (Sora, Veo, Gemini Omni, Nova Reel), speech and audio (TTS, transcription, realtime voice) and sound effects. Every date comes from the provider's own deprecation page, changelog or cloud lifecycle page. It is for engineers, product owners and AI governance teams who must plan migrations and creative workflow changes before a shutdown. Providers are listed alphabetically; order is never a ranking. Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.

Key takeaways

• The biggest media shutdowns of 2026 have already happened. OpenAI shut down the DALL-E 2 and DALL-E 3 API on 12 May 2026, closed the Sora web and app experience on 26 April 2026 and shut down the Sora API and Videos API on 24 September 2026 with no one-to-one replacement.
• More image ids go before February 2027. OpenAI retires gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest on 1 December 2026 and gpt-image-2 on 10 February 2027, pointing to gpt-image-2.6-sunburst or gpt-image-2.6-flare.
• Google consolidated image and video on Gemini models. On the Gemini API, Imagen 4 shut down on 17 August 2026 with gemini-nano-banana-2.1 named as the replacement, and the Veo 3.1 preview ids closed on 22 October 2026 with gemini-omni-1.1-flash named as the replacement.
• The same model has different dates on different platforms. gpt-image-1.5 retires on 1 December 2026 on the OpenAI API and on 16 December 2026 on Azure OpenAI; Veo 3.1 GA on Google Cloud has a minimum retirement of 4 May 2027 while its Gemini API previews are already closed.
• Speech and voice are the next wave. OpenAI's tts-1, tts-1-hd and gpt-4o-mini-tts go on 6 January 2027, its realtime and audio families on 20 January 2027 and whisper-1 on 26 February 2027.

Status key

Shut down in 2025 and 2026

As of 2026-11-15. Providers alphabetical, then by shutdown date. Dates apply to the named platform.

The Sora shutdown reached other platforms too: ElevenLabs removed Sora 2 and Sora 2 Pro from its Image and Video model picker on 24 September 2026 and pointed users to another video model such as Gemini Omni 1.1 Flash. For the background, timeline and data export route, see OpenAI Sora Shutdown Explained.

Retiring from 15 November 2026

As of 2026-11-15. Providers alphabetical, then by shutdown date.

For the GPT Image family itself, its versions and what the sunburst and flare variants are, see OpenAI GPT Image Models Explained.

Same model, different dates: platform comparison

Cloud resellers run their own lifecycles. A team that calls a model through a cloud platform must track that platform's date, not the model maker's. As of 2026-11-15.

Two practical consequences follow. First, a cloud date can be earlier than the maker's own date, as with whisper on Azure. Second, the stated replacement can differ: Google Cloud names veo-3.1-generate-001 as the upgrade for retired Veo 3.0 GA ids, while the Gemini API names gemini-omni-1.1-flash for the same family.

Notice policies by provider

As of 2026-11-15. Providers alphabetical.

Black Forest Labs' own documentation, checked on 2026-11-15, presents FLUX 3 as the latest family, states FLUX.2 remains fully supported and recommends FLUX.2 for new projects on its FLUX.1 Kontext pages; it lists no dated retirements. For model background see FLUX Image Models Explained.

What the shutdowns mean for creative and product teams

• Output style changes: a replacement model is not a drop-in copy. Prompts, aspect ratios, seeds and edit masks tuned for DALL-E 3 or Imagen 4 will produce different images on GPT Image or Nano Banana. Re-run a reference prompt set before cut-over.
• No replacement at all: OpenAI named no one-to-one replacement for the Sora API, so teams had to pick another provider's video model and rebuild the integration, including job polling, storage and moderation handling.
• Stored media: app shutdowns raise an export question. OpenAI's Sora help article points users to an export route and says to export as soon as possible, without giving a fixed deadline.
• Provenance carries over unevenly: the C2PA manifests or watermarks a new model applies may differ from the old one. Check what the replacement marks before you rely on it for labelling. See the AI Content Provenance and C2PA Guide and SynthID vs C2PA Content Credentials.
• Voice products: the speech retirements touch call centres and voice agents. Replacing tts-1 with a realtime model changes the API surface, latency profile and pricing basis, so it is a re-integration rather than an id swap. See the Voice AI Agent Security and Fraud Guide.

How to use this tracker

1. Inventory callers. Search code, configuration and gateway logs for every media model id in this page, including dated snapshots and aliases such as chatgpt-image-latest.
2. Record the platform. Note whether each caller uses the maker's API or a cloud platform, and take the date from that platform's row.
3. Set alerts. Flag any id within 90 days of its shutdown, and subscribe to the deprecation pages and Azure Service Health or Bedrock lifecycle fields that apply.
4. Test the replacement. Run a fixed prompt and asset set through the named replacement, compare outputs, cost per image or second, latency and provenance marks.
5. Cut over in stages. Route a share of traffic to the new id behind a feature flag, keep the old id as rollback until its shutdown date, then remove it.
6. Clean up access. Remove API keys, service accounts and gateway routes that only served the retired model. See the AI Model Retirement Guide.

Coverage notes

This page lists only retirements with a date on an official provider page fetched on 2026-11-15. Midjourney, Runway and Stability AI's own API are not listed because no dated retirement could be confirmed on their official documentation at that check. Language model retirements are in the LLM Deprecation and Retirement Tracker.

Practitioner checklist

• List every image, video, speech and music model id in use, with platform and shutdown date.
• Check the cloud platform's own date where you call a model through Azure, Google Cloud or Bedrock.
• Alert on any id within 90 days of shutdown.
• Run a reference prompt set through the stated replacement before switching.
• Confirm the replacement's provenance marks (C2PA, SynthID, visible labels) still meet your labelling duties.
• Export stored media from consumer apps before they close.
• Revoke keys and routes used only by retired models.

Standards and references

• OpenAI: Deprecations
• OpenAI Help Center: Sora discontinuation
• Google AI for Developers: Gemini API deprecations
• Google Cloud: Model versions and lifecycle
• Microsoft Learn: Model retirement schedule
• Microsoft Learn: Azure OpenAI model deprecations and retirements
• Amazon Web Services: Amazon Bedrock model lifecycle (legacy)
• ElevenLabs: Developer changelog

Related NHI Mgmt Group resources: LLM Deprecation and Retirement Tracker · OpenAI Sora Shutdown Explained · AI Model Retirement Guide · AI Model Provenance and Watermarking Guide