GPT-5.6 Luna — Specs, Pricing & Limits

GPT-5.6 Luna is the cheapest model in its family, and the one detail worth knowing before assuming "cheapest" means "most restricted" in every sense: Luna sits in a genuinely different rate-limit group from Sol and Terra, and the difference runs the opposite direction you'd guess.

Context window and maximum output length match the rest of the GPT-5.6 family exactly — Luna doesn't trade window size away for its lower price, the same way Terra doesn't. Where it actually differs from its siblings is throughput: Luna's batch-queue ceiling and its top-tier requests-per-minute and tokens-per-minute limits are substantially higher than Sol's or Terra's, not lower, at every tier where the two groups diverge. That's counterintuitive if you assume price and throughput move together — here, the cheapest model in the family is also the one with the most generous ceiling once you're spending enough to reach the higher usage tiers.

That makes Luna worth a serious look for any high-volume, latency-tolerant workload — bulk classification, large-scale extraction, anything running enough volume that the rate-limit ceiling itself becomes the binding constraint rather than the per-token price. It's a poor fit for anything that specifically needs Sol or Terra's reasoning quality on a hard task; Luna's position in the family is about price and throughput, not about matching the larger models' capability.

The published long-context repricing rule applies to Luna the same way it applies to the rest of the GPT-5.6 family — crossing the threshold reprices the full request, not just the portion above the line.

The table below is computed directly from this site's facts module: pricing across every published service tier, the long-context rule, and Luna's own rate-limit table.