OpenAI removes o1, o3-mini, o4-mini and GPT-4 Turbo on 23 October. The replacements it names run from 5.5 times the price to 97 percent less.

OpenAI's deprecations page lists a shutdown on 23 October 2026 for eleven models and six fine tuned models. It was announced on 22 April, which is six months, the minimum notice OpenAI's own policy sets for generally available models. The page says that at the shutdown date the model "will no longer be accessible", and it names a replacement for each one.

To compare the landings we priced one reference month: 10 million input tokens and 2 million output tokens, no caching, standard processing, using the list prices on OpenAI's own model pages on 19 October. Your mix will differ, so read the multipliers as direction and not as a quote.

GPT-5.6 Sol costs $4 input and $20 output per million tokens, Terra $2 and $12, and Luna $0.20 and $1.20. OpenAI describes Terra as roughly the mini tier of earlier GPT-5 families and Luna as the nano tier. Each lists a maximum of 922,000 input tokens, and each prices a prompt over 272K input tokens at 2x input and 1.5x output for the whole request. Cached input is a tenth of the input rate, and cache writes bill at 1.25 times the uncached input rate.

If you are coming from o3-mini or o1, whose context window is 200,000 tokens, the 272K rule matters only once you start sending longer prompts. It is worth knowing before you do, because it reprices the whole request and not just the part over the line.

GPT-3.5 Turbo is the clearest case. OpenAI names Terra, which on the reference month costs 5.5 times as much as gpt-3.5-turbo ($44.00 against $8.00). GPT-5.6 Luna would cost $4.40, 0.55 times as much, and gpt-6-luna $2.00, 0.25 times as much. Both are priced on the same OpenAI pages.

GPT-4.1 nano is the second. The named landing, GPT-5.6 Luna, is 2.4 times the price ($1.80 to $4.40), while gpt-6-luna costs $2.00, 1.1 times. OpenAI itself names gpt-6-luna as the replacement for GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini in its 14 October notice, which removes those on 14 January 2027, so it is not an exotic choice.

Price is one axis. A cheaper model that fails your evals is not cheaper, and we have no data on how your workload behaves on any of these. The point is narrower: the mapping table is OpenAI's default and not a cost optimisation, so test the cheaper landing before you accept the named one.

The Sol model page says its price is a promotion: "GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026." It describes $4 and $20 as a 20% cut in input price and a 33% cut in output price, which implies earlier prices of about $5 and $30.

That matters because Sol is the landing for most of the removed models. On the reference month Sol costs $80.00 at $4 and $20, and about $110.00 at $5 and $30. At the higher price the saving against o1 is 59 percent, against o1 pro 96 percent, against GPT-4 74 percent and against GPT-4 Turbo 31 percent. For o3-mini the cost multiple rises from 4.0 to 5.6. The page promises the lower price only through 21 November, so a budget that assumes $4 and $20 for good is a bet.

Sol, Terra and Luna are reasoning models, and so are o1, o3-mini and o4-mini. OpenAI's reasoning guide says reasoning tokens are not visible through the API, occupy space in the context window, and are billed as output tokens. Output is the expensive side of every price above, so the same prompt can cost more or less on a new model depending on how much it thinks.

The guide says GPT-5.6 defaults to a reasoning.effort of medium, and the Terra and Luna pages list none, low, medium, high, xhigh and max. For a workload you are moving off a cheap model, test the lowest effort that passes your evals before you test the default, and compare total output tokens per request and not only the price per token.

The table lists six fine tuned models for removal on 23 October: ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano, ft-babbage-002, ft-davinci-002 and ft-o4-mini. For each it names a recommended replacement base model: Terra, Sol, Luna, Terra, Terra and Terra.

OpenAI's model pages list fine tuning as not supported on gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna and gpt-6-luna. So the replacement base model cannot be fine tuned, and whatever behaviour you trained in has to be rebuilt another way: a longer prompt, examples in context, retrieval, or a different provider. Start with the prompt, since that is the part you can test this week.

The self serve fine tuning notice on the same page narrows the runway further. Organisations that had not run inference on a fine tuned model in the past 60 days could no longer create jobs from 2 July, and active customers lose the ability to create new jobs on 6 January 2027. Inference on a fine tuned model stops when its base model is deprecated.

gpt-image-1 is replaced by gpt-image-2.5-sunburst or gpt-image-2.5-flare. On image tokens gpt-image-1 lists $10 input, $2.50 cached input and $40 output per million tokens, and gpt-image-2.5 lists $8, $2 and $30, which is 20 percent less on input and 25 percent less on output. Text input is $5 on both.

The 2.5 page also says its token rates match GPT Image 2 and that the GPT Image 2 calculator does not estimate GPT Image 2.5 token consumption, so the cost of an image cannot be read off the rate card. Generate a representative batch and read the usage before you commit. The next image removals follow on 1 December, for gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest.

OpenAI documents an optional shutdown_date field on the List models endpoint, a YYYY-MM-DD string or null when none is announced, also returned by Retrieve a model. One request shows every model your key can see that has a date. That lists what is available to the key and not what your code calls, so it complements a code search and does not replace it.

Escalation paths hide most often. Teams route to a reasoning model for the hard case, the retry after a cheaper model failed validation. That branch is the least exercised in testing and the first to break on Friday.

This is the first of several dated removals: older GPT-5 and o3 snapshots on 11 December, GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini on 14 January 2027, and gpt-5.3-codex, gpt-5.1 and gpt-5.4-nano on 1 April 2027. We keep every dated deadline from OpenAI, Anthropic, Google and GitHub on one page, with a calendar feed you can subscribe to.