GPT-5.4 vs GPT-5.6 Terra

A generation apart in name, but the practical gap between these two is mostly about recency and price rather than raw capability — both publish the same window, the same rate-limit group, and the same long-context repricing rule.

What's genuinely different

Terra's knowledge cutoff runs several months ahead of GPT-5.4's, which is the more current model of the two by a meaningful margin. GPT-5.4, in exchange, costs somewhat more per token than Terra across most published tiers — an unusual direction for an older model, and worth confirming against the live tables before assuming "newer costs more" as a blanket rule on this roster, because it doesn't hold uniformly here.

The caching wrinkle

GPT-5.4 doesn't charge for cache writes at all; Terra, as a GPT-5.6-family model, does. For a workload that writes many cache prefixes without reusing most of them, that tips the practical economics back toward GPT-5.4 even though Terra's headline per-token price is lower.

How to actually decide

If your workload's caching pattern reuses prefixes heavily, Terra's lower headline price and more recent knowledge both point the same direction. If it writes a lot and reads back little, run both through the prompt-caching savings calculator before assuming the lower sticker price wins — the write charge can erase Terra's price advantage faster than the raw per-token numbers suggest, and that's the one place this comparison actually needs arithmetic rather than a glance at the table.