GPT-6 Luna pricing and cost calculator
GPT-6 Luna pricing is simple on the surface: $0.10 per 1M input tokens, $0.01 for cached input, $0.125 for cache writes, and $0.50 for output on Standard processing. The part that surprises people is the 272K line. Cross it and the whole request jumps to double rates.
$0.10 in / $0.50 out per 1M tokens
That is the Standard short-context rate for GPT-6 Luna (model ID gpt-6-luna), checked against OpenAI's published pricing on October 13, 2026. Cached input costs $0.01 per 1M; writing to the cache costs $0.125 per 1M. Above 272,000 input tokens, every rate in the request roughly doubles.
GPT-6 Luna cost calculator
Enter your token counts. The calculator picks the right price band automatically and shows the math line by line.
Estimated cost at short-context rates.
| Token type | Tokens | Rate / 1M | Cost |
|---|
Rates are OpenAI's published Standard prices per 1M tokens. The 272K band is judged on input tokens only (uncached + cached + cache writes). Batch and Flex halve the Standard rate; Fast doubles it. Regional uplifts, tools, and reasoning-token accounting are not included.
The full price table
OpenAI publishes two bands for Luna. Short context covers requests with 272,000 input tokens or fewer. Long context starts above that. Rates are USD per 1M tokens on Standard processing.
| Token type | Short context (≤ 272K in) | Long context (> 272K in) |
|---|---|---|
| Input | $0.10 | $0.20 |
| Cached input | $0.01 | $0.02 |
| Cache write | $0.125 | $0.25 |
| Output | $0.50 | $0.75 |
For comparison, the rest of the GPT-6 family on the same terms:
| Model | Input | Cached in | Output |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6 Sol | $2.00 | $0.20 | $10.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
Luna is OpenAI's cheapest current-generation model: one twentieth the input price of Sol, one hundredth of Astra. GPT-6.1 Sol keeps Sol's headline rates but halves the cached-input price to $0.10.
The 272K line that changes the whole bill
OpenAI's long-context band does not work like a tax bracket. When your input passes 272,000 tokens, the higher rates apply to the entire request, not just the tokens above the line. A 273K-token request bills every token at long-context rates.
Three things about the threshold are worth knowing:
- It counts input tokens only. Output tokens do not push you over the line.
- Cache reads and cache writes count toward it. A request with 200K fresh tokens plus 100K cached tokens is over the line.
- Reasoning tokens are billed as output, and they occupy context like any other token.
The practical move: if your workload sits near 272K, split the request or trim the prompt. Crossing the line by 1,000 tokens roughly doubles the input bill. The calculator above flags the band change and shows what the same request would cost under short-context rates.
Prompt caching: writes cost before reads pay
Luna's cheapest token is a cached one: $0.01 per 1M, one tenth the fresh-input price. But the cache is not free to build. Writing a token to the cache costs $0.125 per 1M, which is 1.25 times the fresh-input rate.
The rule of thumb: caching pays off when a prefix is reused enough times that the cheap reads outweigh the one expensive write. For a prefix reused N times, the average input cost per use is roughly ($0.125 + N x $0.01) / N per 1M. At 10 reuses that is about $0.0225 per 1M, less than a quarter of the fresh rate. At 2 reuses it is $0.0725, still cheaper, but the margin is thin.
OpenAI now ships a Prompt Caching Dashboard so you can see real hit rates instead of guessing. GitHub reported that better caching cut the share of prompt tokens needing fresh processing by more than half across billions of Copilot requests, which shows how much of a real agentic bill is cacheable.
Batch, Flex, and Fast
Processing mode multiplies the Standard rate in the same context band:
- Batch: 50% off. For work that can wait up to 24 hours.
- Flex: 50% off. Cheaper tier for non-urgent traffic with the standard API shape.
- Fast: 2x the Standard rate. For latency-sensitive calls.
On Batch, Luna output drops to $0.25 per 1M tokens. At that price, a million output tokens cost less than a cup of coffee in most cities. Note that some modes have availability limits: Fast, for example, is not offered with EU data residency on Astra, so check the pricing page before designing around it. Eligible regional processing adds a 10% uplift on recent models.
Luna vs GPT-5.6 Luna: half the price, slightly less smart
On September 22, 2026, OpenAI replaced the GPT-5.6 tier with GPT-6 Sol and Luna and cut both prices in half. Luna's input rate fell from $0.20 to $0.10 per 1M tokens; output fell from $1.20 to $0.50.
The price cut is real. The capability story is mixed. Independent evaluator Artificial Analysis found GPT-6 Luna scored 2 points lower than GPT-5.6 Luna on its Coding Agent Index, with small regressions on SWE-Atlas-QnA (44% vs 49%) and DeepSWE v1.1 (64% vs 66%). Cost per task fell about 60% at the same time. Luna is the cheaper option for high-volume work. For complex coding, expect slightly weaker results than the model it replaced.
One migration note: there is no GPT-6 Terra. The middle tier between Sol and Luna was dropped, so three-tier routing setups collapse to two.
When Luna makes sense
Luna fits work where volume matters more than peak reasoning: classification, extraction, rewriting, simple Q&A, and first-pass triage. At $0.10 per 1M input, a million-token document sweep costs ten cents on Standard, five cents on Batch.
Pick Sol ($2/$10) when the task needs real coding or agentic ability. Pick Astra ($10/$50) when errors are expensive and stronger reasoning is worth the premium. The honest test is cost per completed task, not cost per token: a cheaper model that needs three retries is not cheaper.
Frequently asked questions
How much does GPT-6 Luna cost per 1M tokens?
$0.10 per 1M input tokens, $0.01 per 1M cached input, $0.125 per 1M cache writes, and $0.50 per 1M output tokens, on Standard processing with 272K input tokens or fewer.
What happens when a request goes over 272K input tokens?
The entire request moves to long-context rates: $0.20 input, $0.02 cached input, $0.25 cache write, $0.75 output per 1M. Only the input side counts toward the threshold; output tokens do not trigger it.
Is GPT-6 Luna cheaper than GPT-5.6 Luna?
Yes. OpenAI cut both rates by 50% on September 22, 2026. Input went from $0.20 to $0.10 per 1M; output went from $1.20 to $0.50.
Is GPT-6 Luna available on the free ChatGPT plan?
Yes. GPT-6 Luna rolled out to ChatGPT Free and Go users on October 8, 2026. The API model gpt-6-luna has been available since September 22, 2026.
How much do Batch and Flex cost?
Both cost 50% of the Standard rate in the same context band. Fast processing costs twice the Standard rate. Availability varies by endpoint: check OpenAI's pricing page before building around a specific mode.
Did GPT-6 Luna get smarter, or just cheaper?
Mostly cheaper. Artificial Analysis measured a 2-point drop on its Coding Agent Index versus GPT-5.6 Luna, with regressions on two coding evaluations, while cost per task fell roughly 60%. For non-coding volume work the tradeoff is clearly worth it.
Should I use Luna, Sol, or Astra?
Use Luna for high-volume simple tasks, Sol ($2/$10 per 1M) for coding and agents, Astra ($10/$50 per 1M) when maximum reasoning quality justifies the price. Measure cost per finished task, not per token.
Sources. Prices are OpenAI's published rates, cross-checked across multiple independent write-ups on October 13, 2026: glbgpt.com's GPT-6 Astra pricing guide (272K band mechanics), explainx.ai's Sol and Luna launch post (price cuts, Artificial Analysis benchmarks, HN reactions), omidsaffari.com's GPT-6 overview (model-by-model rates), and CloudZero's Astra pricing analysis (cache-write ratio, regional uplift). OpenAI's own pricing page is the authority if any number here has moved since.
Prices change. This page was last updated on October 13, 2026. Nothing here is affiliated with OpenAI.
Machine-readable summary: llms.txt
Ask an AI about this guide: