OpenAI schedules GPT-3.5, GPT-4 and GPT-4-turbo shutdowns for July and October

OpenAI published a deprecation notice retiring its legacy model lines — GPT-3.5, GPT-4 and GPT-4-turbo — with hard shutdowns set for July 23 and October 23, 2026. The move effectively closes the GPT-4 era, forcing applications still pinned to those endpoints onto GPT-5.x tiers. Notably, O3 is spared near-term deprecation, a carve-out developers seized on as evidence of which models OpenAI considers strategically durable.

Alongside the deprecations, OpenAI tweaked GPT-5.5 Instant for more natural conversational pacing and fewer bullet-heavy outputs, and discontinued Canvas within that tier. These are product-polish changes, but they arrive bundled with the migration pressure, reinforcing a sense that OpenAI is consolidating users onto its newest, most monetizable stack.

The community reaction skewed negative. Developers on OpenAI's forum and Hacker News voiced frustration over forced, costly migrations — re-testing prompts, re-validating outputs, and absorbing pricing changes on tight timelines. With only O3 exempt, several warned the pattern deepens lock-in to whatever OpenAI deems current, a recurring complaint as the industry churns model versions faster than enterprises can re-certify them.

Competitively, the timing is pointed: as DeepSeek slashes prices and Anthropic cuts Opus costs, OpenAI is pruning cheap legacy options rather than competing on them. That bets enterprises will pay up for capability rather than flee to rivals. What to watch is migration friction between July and October — whether large customers negotiate extensions, and whether the deprecation accelerates multi-provider gateway adoption as teams hedge against future forced moves.