OpenAI retires gpt-5.4-cyber on 20 days' notice, where its policy sets months

OpenAI retires gpt-5.4-cyber on 20 days' notice, where its policy sets months

OpenAI removed gpt-5.4-cyber from its API on October 1, 20 days after it announced the shutdown on September 11. Its own deprecations page sets a minimum notice period of six months for generally available models and three months for specialised variants, a group it defines as chat variants such as gpt-5.1-chat-latest, Codex variants and deep research variants.

The ChatGPT logo and wordmark on OpenAI's blue and orange gradient

The ChatGPT logo and wordmark on OpenAI's blue and orange gradient. Image: OpenAI

The announcement itself runs to two sentences: "The gpt-5.4-cyber model is deprecated and will be removed from the API on October 1, 2026. Migrate to the most capable cyber model available to you before the shutdown date." OpenAI's pricing page and model catalogue no longer list the model.

The policy has exceptions, and OpenAI has not said whether it used one

Those minimums are not absolute. OpenAI prefaces them with "unless safety or compliance concerns require a faster timeline", and adds that if such concerns force an early retirement it will "provide as much notice as reasonably possible". A third category moves faster still: preview models, which OpenAI says are "identified by preview in the model name", can go on about two weeks' notice.

gpt-5.4-cyber does not carry preview in its name, so by OpenAI's own test it is not in that group. Nothing published says which of the other two categories it belonged to, or whether the safety and compliance clause was invoked. What is published is the 20-day gap and the minimums it falls short of, with no explanation attached to either.

Anthropic made a deprecation of its own in the same window, retiring Claude Sonnet 4.5 on 61 days' notice.

What is left in the cyber line, and what it costs

OpenAI's pricing page now lists two cyber models. gpt-5.6-cyber and gpt-5.5-cyber both cost $12.50 per million input tokens, $1.25 per million cached input tokens and $75.00 per million output tokens at short context, with gpt-5.6-cyber also carrying a $15.625 cache-write rate that its older sibling does not list. No price for gpt-5.4-cyber remains published, so there is no way to say from OpenAI's pages what the retired model cost or how the replacements compare against it.

The model catalogue lists GPT-5.6 Cyber, "our most advanced cybersecurity model for authorized vulnerability research and security testing", alongside two aliases. Daybreak Blue is "an alias for flagship general-purpose models with safeguards for defensive cybersecurity work", and Daybreak Red is "an alias for advanced cybersecurity models for authorized vulnerability research and security testing".

Migration advice as thin as the notice

OpenAI names no specific successor in the deprecation entry, only "the most capable cyber model available to you", which leaves a customer to work out what that is for their own account. The catalogue ties these models to authorised vulnerability research and security testing rather than general use, and OpenAI has been extending access through named arrangements, including an offer of its cyber programme to Ukraine in late September.

Those notice periods are what OpenAI asks developers to plan their migrations around. gpt-5.4-cyber got 20 days, and OpenAI has published nothing to say why.

Leave a Comment

Your email address will not be published. Required fields are marked *