GPT-5.6 Comes in Three Tiers: How Sol, Terra, and Luna Actually Differ

OpenAI released GPT-5.6 as three distinct model tiers — Sol, Terra, and Luna — sharing the same API but differing in cost, depth, and latency. Sol, the most capable tier, is priced at $5/$30 per million input/output tokens and is best reserved for high-stakes reasoning tasks where errors are costly. Terra, at $2.50/$15, delivers roughly 97% of Sol's benchmark performance at half the price, making it a practical default for most everyday tasks like coding assistance. Luna, the cheapest and fastest at $1/$6, suits high-volume, repetitive workloads such as classification, tagging, and extraction where per-token cost is the dominant concern. A built-in caching feature — charging only 10% of input price on cached prompt prefixes — can further shift the cost equation, potentially making Terra with caching cheaper per request than uncached Luna.