OpenAI is removing its gpt-5.4-cyber model from the API on 1 October 2026; anything that calls the model by the identifier gpt-5.4-cyber must switch to the named replacement gpt-5.6-cyber before the shutdown date

What this means for you

OpenAI is deprecating **gpt-5.4-cyber** and removing it from the API on **1 October 2026**. This is a targeted, single-model deprecation with a named drop-in replacement — **gpt-5.6-cyber** — so for most callers it is a one-line model-string change rather than an integration rewrite. The people who need to act are anyone who has written `gpt-5.4-cyber` down somewhere: a request that hard-codes the model id, a security or triage pipeline pinned to that model, an eval suite, or a config that selects it by name. After the shutdown date a call naming `gpt-5.4-cyber` no longer resolves to a live model, so those requests start failing until the identifier is updated. The model keeps responding until 1 October, so nothing breaks the day you read this — but the runway is short, and unlike a whole-capability withdrawal there is a direct successor to move to.

Where to go

Before 1 October 2026, replace every reference to `gpt-5.4-cyber` with **`gpt-5.6-cyber`**, OpenAI's named replacement. Grep your code, prompts, CI steps, evals, and any stored configuration for the literal string `gpt-5.4-cyber` — the fix is the same everywhere: swap the identifier. Because it is a like-for-like model successor rather than a capability being removed, there is no data to export and no provider change to plan; the risk is purely a stale model string that stops resolving after the date. Confirm the replacement behaves acceptably for your use case (a newer model can shift outputs), then ship the change ahead of the deadline so nothing errors on 1 October.

• gpt-5.6-cyber — OpenAI's named replacement for gpt-5.4-cyber