GPT-5.4 Cyber

GPT-5.4 Cyber is the earliest known member of OpenAI's GPT Cyber line — variants trained for authorised vulnerability research, exploit validation and penetration testing, and sold only to defenders vetted through the company's Daybreak programme. Almost everything the public knows about it comes from two places where OpenAI could not avoid mentioning it: its price list and its deprecation notice. There was no launch post, no model card and no entry in the public model catalogue; the documentation address that would hold its card returns 404. Internet Archive snapshots show the model appearing in OpenAI's price list between 3 May and 5 June 2026, alongside GPT-5.5 Cyber — but where the newer model carried real rates, the row for GPT-5.4 Cyber showed only dashes in every column, through June, July and August 2026. OpenAI never published what it charged for this model. The end was announced on 11 September 2026: the model is deprecated and leaves the API on 1 October 2026, with GPT-5.6 Cyber named as the migration target. That is twenty days of notice. Across the rest of 2026 OpenAI's published deprecation schedule gives customers roughly six months between announcement and shutdown, which makes this line the sharpest exception in the company's own table. No reason for the short window has been given, and the model has now also disappeared from the price list, leaving GPT-5.5 Cyber and GPT-5.6 Cyber as the only Cyber rows. No capability or safety numbers were ever released for this version. When OpenAI launched GPT-5.6 Cyber on 10 August 2026 it disclosed a comparison with GPT-5.5 Cyber only, and skipped GPT-5.4 Cyber entirely. This profile therefore records what is verifiable — that the model existed, was sold under restricted access, and is being retired — rather than specifications the manufacturer has not published.

▸Videos

No videos yet.

▸Reviews