OpenAI Is Retiring the Entire GPT-4.1 Era: The 2026 Model Graveyard
OpenAI’s developer community deprecation log, updated as of July 14, reveals the full scope of what is being retired in 2026 and what survives. The picture is clear: every GPT-4.x model is being wound down. Only the GPT-5 family remains.
The Models That Survive
OpenAI’s planned persisted model set — the endpoints that will remain live after the deprecation wave completes — is structured around three GPT-5 families:
GPT-5.6 family
gpt-5.6-sol(aliased togpt-5.6)gpt-5.6-terragpt-5.6-luna
GPT-5.5 family
gpt-5.5-2026-04-23(aliased togpt-5.5)gpt-5.5-pro-2026-04-23(aliased togpt-5.5-pro)chat-latest
GPT-5.4 family
gpt-5.4-2026-03-05(aliased togpt-5.4)gpt-5.4-mini-2026-03-17(aliased togpt-5.4-mini)gpt-5.4-nano-2026-03-17(aliased togpt-5.4-nano)gpt-5.4-pro-2026-03-05(aliased togpt-5.4-pro)
No GPT-4.x model appears on this list.
What Is Going Away and When
The deprecation of GPT-4.1-nano fine-tuning is the near-term deadline. Fine-tuned versions of that model will be shut off in approximately three months from the July 14 notice — meaning around October 2026. New fine-tuning runs can still be initiated before that date; inference on existing fine-tuned models becomes unavailable after.
GPT-4.1 and GPT-4.1-mini follow the same trajectory. Neither appears in the persisted set.
The Migration Math
For developers currently on GPT-4.1 (the general-purpose entry model OpenAI pushed in early 2026), the obvious landing spot is GPT-5.4-mini or GPT-5.4-nano — the lowest-cost surviving endpoints in the GPT-5 tier. Pricing on GPT-5.4-nano runs below $1/M for input, making it the closest substitute on cost.
For developers who fine-tuned on GPT-4.1-nano specifically, the migration path is more disruptive. Fine-tuning data and training runs do not transfer directly between model families. The evaluation cost alone — reproducing the fine-tuning dataset against a new base model, validating outputs, retuning hyperparameters — is nontrivial for any production deployment.
The three-month window on fine-tuning shutoffs is short relative to typical enterprise deployment cycles. A team that made GPT-4.1-nano fine-tuning central to a production workflow in Q1 2026 is now looking at a Q4 migration under time pressure.
What the Deprecation Cadence Tells You
OpenAI shipped GPT-4.1 in early 2026 and is retiring the family inside twelve months. GPT-5.4, which launched in March 2026, is staying — but its own tenure on the survival list signals that it is the floor, not a long-term destination. GPT-5.6 Sol is the active intelligence ceiling.
The pattern: OpenAI is treating model families as roughly six-to-nine-month products before the frontier moves on. Developers building on API endpoints with anything shorter than a multi-year horizon need to either target the chat-latest alias or build in migration infrastructure as a first-class engineering concern.
The chat-latest alias is notable precisely because it appears in the persisted set without a date suffix — suggesting it remains a stable pointer regardless of which underlying model it maps to. For latency-insensitive applications where staying on the latest model is acceptable, chat-latest is the lowest-maintenance approach.
GPT-4.x Is Finished
OpenAI shipped its first GPT-4 family model in March 2023. The lineage ends in 2026 — thirty-odd months of a single naming convention, covering dozens of variants across chat, vision, fine-tuning, and instruction-following. The retirement is not quiet: it is the largest simultaneous deprecation in OpenAI’s history, clearing the entire pre-GPT-5 API surface.
Developers have the window. The countdown is running.