SunsettingAI Dev Tools

gpt-5.1, gpt-5.3-codex, gpt-5.4-nano

by OpenAI

Announced

2026-10-01

Complete

2027-04-01

Refund window closes

—

Refund status

No refund applicable

On October 1, 2026 OpenAI announced that `gpt-5.1`, `gpt-5.3-codex` and `gpt-5.4-nano` will be removed from the API on April 1, 2027, six months after the notice. The deprecations page names two substitutes: `gpt-6-sol` for both `gpt-5.1` and `gpt-5.3-codex`, and `gpt-6-luna` for `gpt-5.4-nano`. Read from OpenAI's model pages on October 4, 2026, the swap is a price rise on input for `gpt-5.1` users, a mixed change for `gpt-5.3-codex` users, and a straight price cut for `gpt-5.4-nano` users. The detail most likely to surprise a pipeline that writes prompts, scripts or captions for AI video: `gpt-5.1` and `gpt-5.4-nano` list `none` as their default reasoning effort, while `gpt-6-sol` and `gpt-6-luna` list `medium` as the default. A call that never set the parameter will behave differently after the model string changes.

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 gpt-5.1, gpt-5.3-codex or gpt-5.4-nano usage dated after April 1, 2027, 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
Re-budget before you switch, because the per-1M-token prices listed on OpenAI's model pages on October 4, 2026 move in different directions. gpt-5.1 is $1.25 input, $0.125 cached input, $10 output. gpt-5.3-codex is $1.75 input, $0.175 cached input, $14 output. gpt-5.4-nano is $0.20 input, $0.02 cached input, $1.25 output. The replacement gpt-6-sol is $2 input, $0.20 cached input, $10 output, plus a separate $2.50 cache-write line. gpt-6-luna is $0.10 input, $0.01 cached input, $0.50 output, plus $0.125 for cache writes. So gpt-5.1 to gpt-6-sol raises the listed input rate by 60% with output unchanged, gpt-5.3-codex to gpt-6-sol raises input by about 14% and cuts output by about 29%, and gpt-5.4-nano to gpt-6-luna halves input and cuts output by 60%.

Migration path

Recommended successor

gpt-6-sol and gpt-6-luna

Why this is the closest fit

They are the substitutes OpenAI names on the October 1, 2026 deprecation rows: gpt-6-sol for gpt-5.1 and gpt-5.3-codex, gpt-6-luna for gpt-5.4-nano. Both list Chat Completions, Responses and Batch as supported endpoints, so the request shape you use today is still available.

What differs from the original

Three things change besides the model string. First, the default reasoning effort: gpt-5.1 and gpt-5.4-nano default to none, and gpt-6-sol and gpt-6-luna default to medium, so set reasoning effort explicitly if you relied on the old default for speed or cost. Second, context: the retiring models list a 400,000-token context window and the replacements list 1,050,000, but on both replacement pages input above 272K tokens is priced at 2x the input and cache rates and 1.5x the output rate, so long-context jobs cost more per token past that line. Third, gpt-5.3-codex is listed as supporting the Responses endpoint only, not Chat Completions; gpt-6-sol supports both, so a codex integration built on Responses can stay on Responses. Fine-tuning is listed as not supported on gpt-5.1, gpt-5.4-nano, gpt-6-sol and gpt-6-luna, so there is no fine-tuned model to carry over on those rows.

Alternatives by use case

You use gpt-5.1 to draft video prompts or scripts with reasoning left at its default

On gpt-6-sol, test once with reasoning effort set to none to match today's behavior and once at the new medium default, then compare output tokens per job before you pick one

You run high-volume captioning, tagging or prompt expansion on gpt-5.4-nano

gpt-6-luna lists lower input and output prices, but its default reasoning effort is medium, not none. Set the effort explicitly or the cheaper rate can be offset by extra reasoning output

You use gpt-5.3-codex for build or automation tooling around a video pipeline

Move to gpt-6-sol on the Responses endpoint you already call; the listed output rate drops from $14 to $10 per 1M tokens

You send very long transcripts or shot lists in one request

Check whether your inputs cross 272K tokens on the replacement. Past that point the replacement pages list 2x input and 1.5x output pricing

What this tool meant

gpt-5.1 is the first GPT-5 generation model on OpenAI's retirement list, about a year after its gpt-5.1-2025-11-13 snapshot. The pattern matches the October 23, 2026 batch: a named replacement covers capability, not the defaults, price units or context rules you tuned around. The reasoning-effort default is the easy one to miss, because nothing in your code changes except the model name, and the behavior still does. Pin every parameter you depend on before April 1, 2027, rather than inheriting whatever the next model ships with.