SunsettingAI Dev Tools

gpt-4-turbo, gpt-4, o1, o1-pro, o3-mini, o4-mini, gpt-4.1-nano, gpt-4o-2024-05-13

by OpenAI

Announced

2026-04-22

Complete

2026-10-23

Refund window closes

—

Refund status

No refund applicable

On April 22, 2026 OpenAI announced a legacy-snapshot batch in which every model loses API access on October 23, 2026. The list on OpenAI's deprecations page: `gpt-4-turbo` (and `gpt-4-turbo-2024-04-09`), `gpt-4` and `gpt-4-0613`, `gpt-4-1106-preview`, `gpt-4o-2024-05-13`, `o1` (`o1-2024-12-17`), `o1-pro` (`o1-pro-2025-03-19`), `o3-mini`, `o4-mini`, `gpt-4.1-nano`, `gpt-3.5-turbo` and `gpt-3.5-turbo-0125`, plus `gpt-image-1`. Fine-tuned models built on `gpt-3.5-turbo`, `gpt-4`, `gpt-4.1-nano`, `o4-mini`, `babbage-002` and `davinci-002` go on the same date. Three substitutes cover the whole batch: `gpt-5.6-sol` for the GPT-4 and o1 family, `gpt-5.6-terra` for `o4-mini` and `gpt-3.5-turbo`, and `gpt-5.6-luna` for `gpt-4.1-nano`. The part the announcement does not spell out is cost, and it cuts both ways. Read from OpenAI's model pages on October 2, 2026, the move is a price cut if you are on GPT-4, GPT-4 Turbo or o1, and a price rise if you are on one of the small models.

Refund flow

1
Not applicable. These are pay-per-token API models with no pre-paid balance tied to a model string, so there is no refund category to open.
2
If you are billed for any of these model strings for usage dated after October 23, 2026, treat it as a billing error rather than a refund request: open a ticket at help.openai.com with the invoice line and the request IDs.
3
The number to check before the date is your per-token cost after the switch. Per 1M tokens, input then output, as listed on OpenAI's model pages on October 2, 2026: gpt-4 $30 / $60, gpt-4-turbo $10 / $30 and o1 $15 / $60 all move to gpt-5.6-sol at $4 / $20. o4-mini and o3-mini at $1.10 / $4.40 move to gpt-5.6-terra at $2 / $12. gpt-3.5-turbo at $0.50 / $1.50 also moves to gpt-5.6-terra. gpt-4.1-nano at $0.10 / $0.40 moves to gpt-5.6-luna at $0.20 / $1.20. The gpt-5.6-sol page labels its price promotional and available at least through November 21, 2026.

Migration path

Recommended successor

gpt-5.6-sol, gpt-5.6-terra or gpt-5.6-luna

Why this is the closest fit

They are the substitutes OpenAI names itself, one per row, on its own deprecations page. The model pages describe them as the flagship, mini and nano tiers of the GPT-5.6 family, which maps cleanly onto what each retiring model was being used for.

What differs from the original

Three things change beyond the model string. First, price, in both directions (see the numbers above): a GPT-4 Turbo workload gets cheaper, while an o4-mini workload pays roughly 1.8x on input and 2.7x on output, and a gpt-4.1-nano workload pays 2x on input and 3x on output. Second, fine-tuning: on October 2, 2026 the model pages for gpt-5.6-sol, terra and luna each list fine-tuning as not supported, while the gpt-4.1-nano, o4-mini and gpt-4 pages list it as supported. If you run a fine-tuned model from this batch, the named substitute is a base model you cannot fine-tune, so the tuned behaviour has to move into the prompt or into a model that still accepts fine-tuning. Third, reasoning: all three replacements are reasoning models whose effort setting defaults to medium, with `none` listed as a supported value. If you are replacing gpt-4-turbo or gpt-4.1-nano, which had no reasoning step, set effort explicitly rather than paying for reasoning tokens you did not ask for. Context also grows: the replacements list a 1,050,000-token window and 128,000 max output tokens, against 128,000 and 4,096 on gpt-4-turbo, and the sol page doubles the input price for prompts over 272K tokens.

Alternatives by use case

You run gpt-4-turbo or gpt-4 in production

gpt-5.6-sol is the named substitute and is cheaper per token on the model pages today. Re-run your evaluation set before switching, since output style will shift, and set reasoning effort deliberately

You use o1-pro

OpenAI names gpt-5.6-sol with reasoning.mode set to pro. o1-pro was Responses API only, so the call shape you already have is the one to keep

You chose o4-mini or gpt-4.1-nano for cost

Meter a representative day of traffic on gpt-5.6-terra or gpt-5.6-luna before October 23 and compare the invoice line, because the per-token rates on these rows go up, not down

You depend on a fine-tuned model in this batch

Plan for the substitute not being fine-tunable. Test whether few-shot examples in the prompt recover enough of the tuned behaviour, and treat a fine-tunable model elsewhere as the fallback

You pinned gpt-4o-2024-05-13 for reproducible output

The pin is the thing being removed. Move to gpt-5.6-sol, pin its exact model string, and re-baseline any stored expected outputs

What this tool meant

This batch retires most of the models that a typical 2024 and 2025 AI app was actually built on. GPT-4 Turbo and o1 were the high-end choices, and o4-mini and gpt-4.1-nano were the low-cost options for steps like classification, captioning and prompt expansion. Retirements usually get framed as an upgrade, and for the expensive models that framing holds up on price alone. For the small models it does not, and that is the lesson worth keeping from this one: a named replacement is a replacement for capability, not a promise about cost or about features like fine-tuning. Read the pricing and feature tables for the model you are moving to, not just the deprecation row, and do it before the date rather than after the first invoice.