GPT-5.6 Family API Pricing: Sol, Terra and Luna Compared

GPT-5.6 is a three-tier family with one shape: Sol, Terra and Luna share the same 1,050,000-token context window and the same billing rules, and differ almost entirely in price. That spread is wide — at a standard workload the family runs from $4.00 to $72.00 a month, an 18x gap between the cheapest and the flagship tier. This guide costs all three tiers from OpenAI’s official list prices, the same verified dataset behind our calculator and each model’s own page.

Disclosure: this article contains no affiliate links. Where CostMyTokens does use affiliate links elsewhere, they never affect order — every table is sorted by price only. Read our disclosure.

The family at a glance

Prices are USD per 1M tokens, official OpenAI API list prices, verified 2026-10-25. The two monthly columns use the same 10M tokens under different mixes.

Source for all rows: OpenAI’s official pricing page (source), cross-checked against our dataset on 2026-10-25. No gateway sells any of the three at a verifiable public price below the official rate, so unlike the open-weight models there is no gateway column — official is the only way to buy them.

Luna: the volume tier

GPT-5.6 Luna is the family’s high-volume tier at $0.2 input / $1.2 output. At the standard mix it is $4.00 a month, and its caching rate is the family’s deepest in absolute terms: $0.02 per 1M cached input, 90% off list. A document pipeline doing 1,000 long calls a month (50K input, 2K output each) runs $12.40 on Luna per its model page — the kind of workload where the tier choice barely matters next to the token count.

One honest caveat: within OpenAI’s own catalog, Luna is not the cheapest way to buy volume. GPT-6 Luna undercuts it on both columns ($0.1 / $0.5, $1.80 at the same mix) — see the full ladder in our OpenAI API pricing overview. GPT-5.6 Luna’s case is generational, not financial: same billing shape, one generation back, for teams pinned to the 5.6 line.

Terra: the middle tier with an output catch

GPT-5.6 Terra sits at $2 / $12 — ten times Luna’s monthly cost at the standard mix ($40.00). The number to watch is the output rate. Terra’s input matches GPT-6.1 Sol’s $2 exactly, but its output is $12 against Sol’s $10, so output-heavy work quietly costs more on Terra: at $90.00, flipping the mix from 8/2 to 3/7 more than doubles its bill on output share alone. Terra’s cached input ($0.2, 90% off) and batch rate (exactly half) are in line with the rest of the family; it is specifically long generations — reports, code, drafts — where Terra’s price position weakens.

Sol: the family flagship, on a promotional clock

GPT-5.6 Sol tops the family at $4 / $20, $72.00 a month at the standard mix and $152.00 output-heavy. Against OpenAI’s frontier tier, GPT-6 Astra ($10 / $50), Sol costs 40% as much on both columns — it is the “one rung down from frontier” price point, and for many production workloads it is where the family’s price/performance argument is strongest.

Two Sol-specific facts belong in any budget. First, OpenAI lists Sol’s current pricing as promotional, available at least through November 21, 2026 — the $4 / $20 rate is not guaranteed beyond that date, and this page’s figures should be re-checked after it. Second, Sol’s agent-loop economics are dominated by caching: its model page works an agent loop with heavy prompt reuse (10K fresh input, 100K cached input, 2K output per call, 500 calls a month) at $60.00 a month — less than its plain standard-mix figure despite moving far more tokens, because 100K of every call’s input bills at the $0.4 cached rate instead of $4.

How the mix moves the ranking

The order inside the family never flips — Luna is always cheapest, Sol always dearest — but the gaps change shape:

• Output share. Flipping 10M tokens from 8/2 to 3/7 multiplies Luna’s bill by 2.25 ($4.00 → $9.00), Terra’s by 2.25 ($40.00 → $90.00) and Sol’s by 2.1 ($72.00 → $152.00). All three output rates are 5–6x their input rates, so generation-heavy products should budget from the output column, not the input sticker.
• Caching. With 6M of 8M input tokens served from cache (plus 2M output), the standard workload costs $2.92 on Luna, $29.20 on Terra and $50.40 on Sol — roughly 27–30% below the uncached figures. The discount rate is 90% on all three tiers, so caching narrows the absolute gap without changing the order. The math is worked through in our prompt caching guide.
• Batch. Every tier’s batch rate is exactly half of standard — $2.00 / $20.00 / $36.00 a month at the standard mix — for work that tolerates a batch window; see batch API discounts explained.
• Long prompts. All three tiers bill prompts above roughly 272K input tokens at 2x input and 1.5x output for the full request, not just the excess. A single oversized prompt can double a call’s cost; our pricing traps guide covers this pattern across providers.

GPT-5.6 or GPT-6?

On price alone, the families interleave rather than stack. GPT-6 Luna beats GPT-5.6 Luna on every column. GPT-6.1 Sol and GPT-5.6 Terra share a $2 input but split on output ($10 vs $12) and cached input ($0.1 vs $0.2 — the 6.1 reads cache at half Terra’s rate). GPT-5.6 Sol undercuts GPT-6 Astra by 60% while sitting one tier above GPT-6.1 Sol. So the 5.6 family is not simply “last year’s prices”: Terra and Sol both hold defensible positions between the GPT-6 tiers, and the full fourteen-model ladder in our OpenAI API pricing overview shows exactly where each one slots in. For the GPT-6 side of the comparison, see the GPT-6 API pricing guide.

Run your own numbers

Every figure above is one fixed mix at list price. Plug your real input, output and cached token counts into the CostMyTokens calculator — or open the Sol, Terra and Luna pages to cost a single tier with its own worked examples.