GPT-5.5 — Specs, Pricing & Limits

GPT-5.5 sits between the GPT-5.4 and GPT-5.6 generations chronologically, and its knowledge cutoff reflects that position directly — newer than GPT-5.4's, older than the current GPT-5.6 family's.

Its context window, rate-limit group and long-context repricing rule all match the current frontier generation exactly, which makes the meaningful difference between this model and its GPT-5.6 counterpart almost entirely about recency rather than capability on paper. For workloads where very recent knowledge doesn't move the outcome — well-established libraries, stable APIs, general reasoning tasks — that narrows the practical gap between choosing this model and choosing the newer one considerably.

One detail worth knowing if you're pricing a caching-heavy workload: no cache-writes price is published for this model at any service tier, which is consistent with it predating the GPT-5.6 family's introduction of a charge for writing into the cache. Caching a prefix here costs nothing extra to write, the same as on every model before the newest generation — a genuinely simpler economics than the model this site most often compares it against.

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.