GPT-5.6 Terra vs GPT-5.6 Luna

Terra and Luna sit next to each other on price within the GPT-5.6 family, but they diverge on something a price table alone won't show you: which rate-limit group they actually belong to.

The window is the same, the throughput isn't

Both models publish the identical context window and maximum output length. Where they split is rate limits — Luna sits in a different published group from Terra, and at the higher usage tiers, Luna's batch-queue ceiling and top-tier RPM/TPM are substantially larger than Terra's, not smaller, despite Luna being the cheaper of the two.

Price moves in the expected direction, at least

Luna costs meaningfully less than Terra across every service tier — that part of the comparison is the ordinary "cheaper model, lower price" pattern the rest of this roster mostly follows. It's the throughput divergence layered on top of that price gap that makes this pairing worth checking closely rather than assuming Terra's numbers.

How to actually decide

For a high-volume, throughput-sensitive workload, Luna's combination of lower price and higher ceiling at the top tiers makes it the stronger pick almost by default — there's no throughput tradeoff being made for the lower price here, unusually. For a task where Luna's answer quality genuinely isn't sufficient, Terra is the mid-tier step up before reaching for Sol specifically, and it's worth trying before assuming the top-priced model in the family is necessary.