GPT-5.4 Cyber Pricing Is Not Published
The mistake
Trying to price a workload against gpt-5.4-cyber — it appears in the full model-id listing on the pricing page's own source, so it's reasonable to assume it's just another selectable, priced model — and finding every price cell for it empty, or the model's own reference page returning nothing at all.
Why this happens
Being listed and being priced aren't the same thing. gpt-5.4-cyber shows up in the raw list of model identifiers the pricing page's source enumerates, which is exactly the kind of detail that only turns up if you're reading page source rather than the rendered table — and its own dedicated model page doesn't resolve to real content either. Both surfaces were checked directly rather than assumed from each other's absence, and both come back the same way: nothing published.
Why it matters
A calculator, script, or cost model that includes this identifier as if it were priceable will either error or, worse, silently treat a missing price as zero or fall back to a different model's rate — both of which produce a confidently wrong number rather than an honest gap. For any tooling built against the full model-id list rather than the filtered, priced subset, this is exactly the kind of entry that needs to be excluded explicitly rather than assumed to behave like everything else on the list.
The fix
Don't select this model in any cost calculation — this site's own tools exclude it from every calculator for exactly this reason. If your own tooling enumerates models from a raw list rather than a curated priced set, filter explicitly against what's actually published rather than assuming every listed identifier is usable, and revisit that exclusion periodically rather than assuming it's permanent — an unpublished model can become priced in a future release without much announcement.
There's a sibling model in the same unpriced position — a GPT-5.5-generation Cyber variant shows the identical pattern, listed by id, unpriced, its own reference page equally unreachable. Whatever this model family is intended for, neither generation of it is ready to be planned against yet on this site or, as far as the checked surfaces show, anywhere OpenAI itself currently publishes.
See also
GPT-5.4 Cyber — Specs, Pricing & Limits records this honestly as not published, with the specific surfaces checked, rather than leaving the entry blank with no explanation.
Verified 2026-08-09 against CodexHow facts module (src/data/facts/) — see /about/#accuracy.