Migrating: GPT-5.4 to GPT-5.6 Sol

Moving from GPT-5.4 to GPT-5.6 Sol is a jump of one generation that the lineup has since overtaken — GPT-6 models now sit above both — so it's worth knowing exactly what this particular move buys before choosing it over the newer options.

What actually moves

The knowledge cutoff jumps forward by several months. Price goes up on input, cached input and output, even at Sol's promotional rates. Caching gets dearer too: GPT-5.4 bills a written token at its ordinary input rate, while Sol bills it at a cache-write rate set above Sol's own, higher input rate. Reasoning changes shape — GPT-5.4 defaults to none, Sol to medium, and Sol adds max — so an unconfigured request starts reasoning after the move.

What stays put

Context window, output ceiling, the rate-limit table and the long-context rule all carry over. A workload sized against GPT-5.4's window and throughput doesn't need re-sizing.

Before switching

Price the move with the prompt-caching savings calculator if the workload caches heavily, and remember that Sol's current prices are promotional, with a published "at least through" date and nothing beyond it.

The options this page didn't used to have

If recency is the motive, GPT-5.6 Terra matches Sol's specs for less, and GPT-6 Sol — newer again — costs less than GPT-5.4 on input, cached input and output while sharing its window and rate-limit table. In Codex with ChatGPT sign-in, GPT-5.4 is already gone, retired on August 31, 2026, and GPT-6 Sol is OpenAI's named replacement there.