Migrating: GPT-5.4 to GPT-5.6 Terra

This is the migration worth checking before defaulting to the more obvious GPT-5.4-to-Sol move — Terra matches GPT-5.4 on window and rate-limit group, and actually costs less per token at most tiers, not more.

What actually moves

Knowledge cutoff jumps forward the same way it would moving to any current GPT-5.6-family model. Price moves down on Standard input and output rather than up — an unusual direction for a "move to the newer model" migration on this roster, and worth double-checking against the live tables rather than assuming the opposite. Terra also introduces the GPT-5.6 cache-write charge, the same as every model in its generation — free writes on GPT-5.4 become a real, if modest, per-write cost on Terra.

What stays put

Context window, maximum output, rate-limit group and the long-context repricing rule all carry over unchanged from GPT-5.4.

What to check before switching

If the only reason to stay on GPT-5.4 was price, this migration removes that reason entirely — Terra is both newer and cheaper. The one thing still worth checking is the caching math: a workload that writes a lot of cache prefixes without reusing them heavily should run the numbers through the prompt-caching savings calculator before switching, since the new write charge is the one cost that moves in the wrong direction even as the headline price moves in the right one.