GPT-5.4 — Specs, Pricing & Limits

GPT-5.4 is the previous generation's frontier model — the one most migration guides on this site are written against, since it's the most common starting point for anyone moving up to the newer GPT-5.6 family.

Its context window and rate-limit group match the current GPT-5.6 frontier models exactly, and it publishes the same long-context repricing rule they do: crossing the threshold reprices the full request, not just the excess. Its knowledge cutoff sits noticeably further back than the newer generation's, which is the single most common reason a migration off this model gets prioritized — recency, more often than price, is what actually motivates the switch.

One pricing detail worth knowing if you're auditing a bill rather than just planning one: this model's Batch-tier cached-input rate is published as a figure that doesn't match what halving the standard cached rate would predict — OpenAI rounds it in the source table, and this site ships the published figure rather than "correcting" it to the number arithmetic suggests. If a script you've built computes that figure independently instead of reading it from a table, it's worth checking against the real published value directly.

Unlike the newest generation, this model doesn't charge for writing into the prompt cache at all — cache writes are free here, the same as on every model that predates GPT-5.6's cache-write charge.

The table below is computed directly from this site's facts module: pricing across every published service tier, the long-context rule, and the rate-limit table.