Wave 2 focuses on legacy chat/reasoning models, compact efficiency models, and image generation models, detailing deprecated models and their replacements.

OpenAI plans a two-wave shutdown of dozens of AI models in 2026, with Wave 1 on July 23 and Wave 2 on October 23, followed by further cuts into early 2027 that could affect chatbots, voice assistants, image generators, and coding tools.

Migration guidance urges locating and replacing hardcoded model references, testing replacements, starting fine-tuning migration early, and tracking the July 23 deadline to avoid last‑minute failures.

The piece warns of risks to end-user applications when underlying models are retired, urging organizations to audit their tech stacks and migrate early to prevent service disruption.

Wave 1 targets categories such as search and real-time conversation tools, audio/voice tools, coding tools, and deep research tools, listing deprecated models and recommended replacements.

For non-developers, the piece explains the basics of AI models, APIs, and the difference between deprecated and legacy to clarity the shutdown’s impact.

Fine-tuned models are particularly vulnerable since customizations don’t automatically transfer to new base models, with migration paths specified for several fine-tuned families.

OpenAI provides a deprecations page and developer forums as resources, emphasizing prompt migration to minimize disruption to AI-powered products.

A two-wave shutdown timeline is outlined: Wave 1 on July 23, 2026 (18 models) and Wave 2 on October 23, 2026 (11 more models plus fine-tuned families), with further changes to fine-tuning through January 6, 2027.

Fine-tuning access will tighten progressively: new organizations cannot create jobs after May 7, 2026; inactive organizations lose access after July 2, 2026; active users lose the ability to create new fine-tuning jobs after January 6, 2027.