What GPT-5.6 Sol, Terra and Luna Cost
GPT-5.6 Sol costs $4.00 per million input tokens and $20.00 per million output. Terra costs $2.00 and $12.00. Luna costs $0.20 and $1.20, which makes it the cheapest tier in OpenAI's current line.
The Sol, Terra and Luna names replaced the older Instant, Thinking, mini and nano labels in July 2026. Most material on "gpt-5.6 pricing" predates that rename.
Every number on this page was checked on 16 September 2026 against OpenAI's and Anthropic's own published pricing documentation.
| OpenAI model | Input / 1M | Output / 1M | 1M in + 1M out |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | $60.00 |
| gpt-5.6-sol | $4.00 | $20.00 | $24.00 |
| gpt-5.6-terra | $2.00 | $12.00 | $14.00 |
| gpt-5.6-luna | $0.20 | $1.20 | $1.40 |
| gpt-5.5 | $5.00 | $30.00 | $35.00 |
| gpt-5.4 | $2.50 | $15.00 | $17.50 |
| gpt-5.4-mini | $0.75 | $4.50 | $5.25 |
| gpt-5.4-nano | $0.20 | $1.25 | $1.45 |
The right-hand column is not a published price. It is arithmetic at the listed rates.
gpt-6-astra is the current flagship, launched in early September 2026, and it is the only model here with a published long-context tier: $20.00 / $75.00 above 200k tokens.
AI Perks tracks $7.7M in credits across 194 companies, which is the other half of this arithmetic.
The Price Gaps Between Sol, Terra and Luna
The step from Terra down to Luna is exactly 10x on both input and output. The step from Sol down to Terra is 2x on input but only 1.67x on output.
That asymmetry decides more than the headline numbers do. Ratios below are derived from the listed prices.
| Step down | Input | Output | Blended 1M + 1M |
|---|---|---|---|
| Sol to Terra | 2.0x | 1.67x | 1.71x |
| Terra to Luna | 10.0x | 10.0x | 10.0x |
| Sol to Luna | 20.0x | 16.7x | 17.1x |
| Astra to Sol | 2.5x | 2.5x | 2.5x |
A second pattern in the same arithmetic: Sol charges 5x its input rate for output, while Terra and Luna both charge 6x. The two cheaper tiers penalise verbose generation hardest in relative terms.
The practical consequence: on Terra and Luna, prompt size is rarely what moves your bill. Generated length is. Trimming a system prompt saves less than capping max_tokens.
How GPT-5.6 Compares to Claude and Meta Muse
Sol is exactly 2x Claude Sonnet 5 on both input and output. Terra matches Sonnet 5's $2.00 input price exactly and charges 20% more for output. Luna is the cheapest line in the table by a wide margin.
| Model | Input / 1M | Output / 1M | 1M in + 1M out |
|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $60.00 |
| Claude Mythos 5.1 | $10.00 | $50.00 | $60.00 |
| gpt-6-astra | $10.00 | $50.00 | $60.00 |
| Claude Opus 5 | $5.00 | $25.00 | $30.00 |
| gpt-5.6-sol | $4.00 | $20.00 | $24.00 |
| gpt-5.6-terra | $2.00 | $12.00 | $14.00 |
| Claude Sonnet 5 | $2.00 | $10.00 | $12.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $6.00 |
| Meta Muse Spark 1.3 | $1.25 | $4.25 | $5.50 |
| gpt-5.6-luna | $0.20 | $1.20 | $1.40 |
Right-hand column: arithmetic at the listed rates, not a published price.
The $10/$50 ceiling is now a convention. gpt-6-astra, Claude Fable 5.1 and Claude Mythos 5.1 are priced identically. Mythos 5.1 is limited availability.
Luna has no peer at the bottom. At $1.40 blended it is 4.3x cheaper than Claude Haiku 4.5 and 3.9x cheaper than Meta Muse Spark, both estimates from the listed prices.
Sonnet 5 got cheaper by not getting more expensive. Anthropic had scheduled Sonnet 5 to rise from $2/$10 to $3/$15 on 1 September 2026, then cancelled the increase. Anthropic's own page states: "The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur." The introductory price is now the standard price, which is what puts Terra and Sonnet 5 level on input.
Meta Muse also sells a Contributor tier at $0.10 / $0.20, where Meta trains on your prompts and completions, and Muse Image at $0.01 per image. Meta runs no startup credit program for Muse, so that bill is paid out of pocket. Credits for the rest are tracked at getaiperks.com.
What Each Tier Costs Per Month at Real Volume
At 50M input and 10M output tokens per month, Luna costs about $22, Terra about $220, Sol about $400 and Astra about $1,000.
That volume is a modest production assistant. All figures are estimates from the listed prices, with no discounts or batch pricing applied.
| Model | Monthly estimate | Annualised estimate |
|---|---|---|
| gpt-6-astra | $1,000 | $12,000 |
| Claude Opus 5 | $500 | $6,000 |
| gpt-5.6-sol | $400 | $4,800 |
| gpt-5.6-terra | $220 | $2,640 |
| Claude Sonnet 5 | $200 | $2,400 |
| Meta Muse Spark 1.3 | $105 | $1,260 |
| Claude Haiku 4.5 | $100 | $1,200 |
| gpt-5.6-luna | $22 | $264 |
The spread is 45x, from $22 to $1,000 a month, for identical token counts.
Run that against credits. A $5,000 grant covers roughly 12 months of Sol, 22 months of Terra, or over 18 years of Luna at this volume. Which tier you pick is the difference between credits lasting a quarter and credits lasting past your Series A. AI Perks covers which grants are available.
Which Tier Fits Which Workload
Pick Luna for volume, Terra as the default, Sol for the hard fraction, and Astra only where Sol demonstrably fails.
Percentages below are derived from the listed prices.
gpt-5.6-luna, $0.20 / $1.20. Classification, routing, extraction, tagging, moderation, reranking, short structured output. Anything you run millions of times where the answer is short and the format is fixed.
gpt-5.6-terra, $2.00 / $12.00. The default production path. It is 20% cheaper than gpt-5.4 on both input and output, so there is no price argument for staying on 5.4.
gpt-5.6-sol, $4.00 / $20.00. The hard fraction of your traffic. Sol is 31% cheaper blended than gpt-5.5, the tier it effectively supersedes, so the same reasoning applies.
gpt-6-astra, $10.00 / $50.00. 2.5x Sol across the board, plus a long-context surcharge above 200k tokens. Reserve it.
The real saving is not picking one. Route 80% of that 50M / 10M workload to Luna and the remaining 20% to Sol and the monthly estimate is about $98 rather than $400, a 76% reduction against sending everything to Sol. A single-model architecture is the most expensive decision on this page.
What Is Not Published, and What Is Being Shut Down
OpenAI has not published long-context threshold pricing for Sol, Terra or Luna. The only tiered-context price in the current line is Astra's $20 / $75 above 200k tokens.
Not claimed here, because not verified:
- Benchmark scores or capability rankings. No MMLU, HumanEval or SWE-bench numbers. Price is verifiable, quality at this date is not.
- Context window sizes, beyond Astra's 200k pricing threshold.
- xAI and Grok token prices. Sources conflict badly. Pricing varies by model, so check xAI's own console.
- Google Gemini prices. Not verified as of 16 September 2026, so they are omitted rather than guessed at.
- Rate limits and free-tier ceilings for any provider.
One shutdown matters this month. The gpt-5, gpt-5-mini and gpt-5-nano APIs shut down on 10 December 2026. Do not start new work on them.
That shutdown also raised the floor, and almost nobody has said so. gpt-5-nano was $0.05 / $0.40. Luna, the cheapest current replacement, is $0.20 / $1.20: 4x the input price and 3x the output price at the listed rates. If your unit economics were built on nano, they need rebuilding before December.
Covering the Bill With Free Credits
The cheapest tier is still not free, and provider credit programs routinely cover more than a year of the monthly estimates above.
AI Perks tracks $7.7M in credits across 194 companies, including the major model providers on this page. An amount that looks small beside Astra's estimated $12,000 a year is close to unlimited runway on Luna at the same volume. Pricing and credit terms both move, so re-check quarterly at getaiperks.com.
Frequently Asked Questions
How much does GPT-5.6 Sol cost?
GPT-5.6 Sol costs $4.00 per million input tokens and $20.00 per million output, verified 16 September 2026. One million input plus one million output comes to $24.00. At 50M input and 10M output per month, that is roughly $400. Credits are tracked at getaiperks.com.
Is GPT-5.6 Luna cheaper than Claude Haiku 4.5?
Yes, substantially. Luna is $0.20 / $1.20 per million tokens against Haiku 4.5 at $1.00 / $5.00. That is 5x cheaper on input, about 4.2x on output, and 4.3x for a million in plus a million out, all derived from the listed prices. Luna is the cheapest current tier from either provider.
Should I use GPT-5.6 Sol or Terra?
Use Terra as your default and Sol for the hard fraction of requests. Sol costs 2x Terra on input but only 1.67x on output, so the premium is smallest on generation-heavy work. In the routing example above, reserving Sol rather than defaulting to it cuts the estimated monthly bill by about three quarters.
Did Claude Sonnet 5 get a price increase in September 2026?
No. Anthropic had scheduled Sonnet 5 to rise from $2/$10 to $3/$15 on 1 September 2026, then cancelled it. Anthropic's page states the increase "will not occur." The $2/$10 price, originally introductory, is now standard, which leaves Sonnet 5 level with GPT-5.6 Terra on input.
Is gpt-6-astra worth 2.5x the price of Sol?
Astra is $10 / $50 against Sol's $4 / $20, with a further $20 / $75 tier above 200k tokens. The price gap is verified; the capability gap at this date is not published, so test both on your own traffic.
What happens to gpt-5, gpt-5-mini and gpt-5-nano?
Those APIs shut down on 10 December 2026, so do not build new systems on them. The replacement floor is also higher: gpt-5-nano was $0.05 / $0.40, while Luna is $0.20 / $1.20. Budget for it, and offset it with credits from AI Perks.
Pick the right tier, then let someone else pay for it.