GPT-5.6 Sol vs GPT-5.6 Terra

Same family, same window, same knowledge cutoff, same rate-limit table — the comparison between these two still comes down to price, but the reasons people used to give for picking Sol over Terra have mostly gone.

What's identical

Context window, maximum output, knowledge cutoff, effort levels (none through max, medium by default), the long-context rule and every rate-limit row match exactly. Neither model offers more headroom or more recent training data than the other.

What differs

Terra is cheaper on every row of the table: input, cached input, cache writes and output, in both context bands and every service tier. Sol's August promotional cut narrowed the gap but didn't close it, and that promotion has a published "at least through" date rather than a permanent one.

What changed around them

The old argument for Sol was inertia — it was the model the Codex CLI ran unless you changed it. That stopped being true when GPT-6.1 Sol became the default. And GPT-6 Sol now sits at Terra's input price with a lower output rate and a newer cutoff, so any decision between these two is worth widening to three.

How to decide

If nothing in your workload has been shown to need Sol specifically, Terra does the same job on identical specs for less. If you do need Sol-level quality, test GPT-6 Sol and GPT-6.1 Sol on the same task before settling on the older Sol — on today's table, both cost less than it.