benchrStat hero shows Sol, the top GPT-5.6 tier, at its post-cut rate. Terra and Luna's prices, the context windows, and the API IDs are all in the table below.
The full price and spec table
| Model | Input | Output | Cached input | Long-context input/output | Context | Max output | API model ID |
|---|---|---|---|---|---|---|---|
| Sol (top tier) | $4.00 | $20.00 | $0.40 | $8.00 / $30.00 | 1,050,000 | 128,000 | gpt-5.6-sol |
| Terra (balanced) | $2.00 | $12.00 | $0.20 | $4.00 / $18.00 | 1,050,000 | 128,000 | gpt-5.6-terra |
| Luna (cheapest) | $0.20 | $1.20 | $0.02 | $0.40 / $1.80 | 1,050,000 | 128,000 | gpt-5.6-luna |
These are the current rates on OpenAI's API pricing page. The series has been repriced twice in under a month: Terra fell 20% (from $2.50/$15) and Luna 80% (from $1/$6) on July 30, then Sol fell 20% on input and 33% on output (from $5/$30) on August 21. The 1.05M context and 128K max output specs never moved.
Compare each tier with its nearest alternative
The cleanest way to sanity-check GPT-5.6's prices is to line each model up against an existing OpenAI model at a price you can verify on the live pricing page.
Sol = the GPT-5.5 sticker. Sol's $5/$30 is identical to GPT-5.5 at $5/$30 and didn't move in the July 30 cut. So the flagship of the new series still carries no headline price increase over the current flagship — you'd be paying the same per token for whatever capability gain Sol delivers. For external reference at that tier, Claude Opus 4.8 is $5/$25 (cheaper output) and Gemini 3.1 Pro is $2/$12 (cheaper on both sides).
Terra now undercuts the GPT-5.4 sticker. Terra used to match GPT-5.4's $2.50/$15 exactly; after the cut it's $2/$12, cheaper than GPT-5.4 on both input and output. OpenAI still frames Terra as GPT-5.5-class capability, so the pitch got stronger: GPT-5.5-grade output below the GPT-5.4 price, not just at it.
Luna now undercuts GPT-5 Mini. Luna used to sit between GPT-5 ($1.25/$10) and GPT-5 Mini ($0.25/$2). After an 80% cut, Luna's $0.20 input is now below GPT-5 Mini's $0.25, though its $1.20 output still sits below Mini's $2.00 — Luna wins on both sides now. For an external floor, DeepSeek V4-Pro is $1.32/$3.96 at peak and $0.66/$1.98 off-peak after its August 16 rise: Luna now beats it on both sides of the rate card at any hour.
The caching and long-context policy
Cached input runs $0.40/$0.20/$0.02 for Sol/Terra/Luna. Long-context pricing (above the 272K-token threshold) is $8/$30 for Sol, $4/$18 for Terra, and $0.40/$1.80 for Luna. Sol also lists cache writes at $5.00 short-context and $10.00 long-context, Batch and Flex at half of standard ($2/$10), and Fast mode — the replacement for priority processing — at double ($8/$40). Terra and Luna haven't had their batch and flex rows restated since July 30, so treat those as unconfirmed. The budgeting point survives every repricing: output dominates heavy agent workloads.
Cost scenarios
Take a steady agent at 30M input + 10M output per month. On Sol that's $120 + $200 = $320/month, down from $450 before August 21; with a 90% cache hit on input it falls to about $12 + $200 = $212/month. The same volume on Terra is $60 + $120 = $180/month, and on Luna $6 + $12 = $18/month. The gap that mattered most just narrowed: Sol used to cost 2.5x Terra on this workload, and now it's 1.8x.
Use-case fit
Consider for: teams already standardized on the OpenAI stack. Terra now beats the GPT-5.4 rate outright while pairing it with provider-reported SWE-bench Verified (85.2%) and GPQA Diamond (88.0%) scores; Luna is the tier to reconsider if you dismissed it before the cut; and Sol still has the line's 1,050,000-token context maximum. Test the selected tier on your own latency, quality, and cost targets.
Skip if: you are already happy on GPT-5.5 at the same $5/$30 sticker and do not need Sol's benchmark gains or newer tool/reasoning behavior. There is no price-only reason to migrate from GPT-5.5 to Sol — that comparison didn't change.
Decision checklist
The July 30 cut is official and live on OpenAI's pricing page, not a preview announcement. If you budgeted for the old Terra/Luna rates, redo the math — Luna in particular is a fifth of its old cost per token.
Terra's "GPT-5.5-class" claim has the same benchmark numbers behind it as before — SWE-bench Verified 85.2% and GPQA Diamond 88.0%, from OpenAI's system card — now at a lower price. Benchmark it against your own workloads either way; a price cut doesn't change whether the quality claim holds on your tasks.