GPT-5.6 Luna: economy for high-volume work
GPT-5.6 Luna is the most economical model of OpenAI's GPT-5.6 series, generally available since July 9, 2026. It is built for data-heavy, latency-sensitive workloads, priced on Genosai at 80 ₽ per 1M input and 480 ₽ per 1M output tokens. Even so, in agentic coding tests Luna beats the Claude Opus 4.8 flagship.
Updated: July 23, 2026
- Price 80/480 ₽ per 1M — One of the most affordable models in the catalog, at current-generation GPT quality.
- Beats Opus 4.8 on code — On the Artificial Analysis Coding Agent Index it outperforms the Claude flagship in about a third of the time.
- Built for volume — Chats, classification, and lightweight agentic workflows — tasks where speed and throughput matter.
- 256,000-token context — A large window for an economy model: a long document fits in one pass.
- Vision and files — Accepts images in JPG, PNG, and WebP formats, up to 10 files per request.
Contents
- What is GPT-5.6 Luna
- Capabilities
- Examples prompt and response
- How to use on Genosai
- Prompts
- Generation cost
- How it compares
- Limitations and tips
- FAQ
What is GPT-5.6 Luna
GPT-5.6 Luna is the smallest model in OpenAI's GPT-5.6 series. The series became generally available on July 9, 2026 and comes in three tiers: the Sol flagship, the balanced Terra, and the economical Luna. Luna targets data-heavy, latency-sensitive workloads: chats, classification, lightweight agentic workflows. In the API documentation it corresponds to the nano tier of earlier GPT-5 generations.
Normally a bottom-tier model comes with a visible drop in quality. Here the situation is different, and that is the real story of the series: OpenAI bet on performance per dollar, and the payoff shows most strongly at the bottom step. On the Artificial Analysis Coding Agent Index, Luna outperforms Claude Opus 4.8 — a rival flagship — in roughly a third of the time, at half the output tokens and about a quarter of the estimated cost.
In knowledge work Luna nearly reaches the peak-performance level of GPT-5.5, OpenAI's previous flagship, at under half the estimated cost. Together with Terra it surpasses Claude Fable 5 at roughly one sixteenth of the cost. That is an unusual set of characteristics for an economy model.
On Genosai, Luna has a 256,000-token context window — the same as Sol and Terra — plus image support. The client price is 80 ₽ per 1M input and 480 ₽ per 1M output tokens. The result is a model you can comfortably put into a pipeline: at volume the savings are real, and quality stays at current-generation level.
Capabilities
GPT-5.6 Luna is designed for throughput: many calls, fast answers, predictable cost.
High-volume tasks and classification
Luna's core scenario is work with many calls, none of them individually hard: tagging and classifying incoming messages, support replies, extracting fields from text, short descriptions and translations. In those tasks what decides the matter is not benchmark records but the cost per thousand calls and the response time. Luna is optimized for exactly that, and at volume the gap against pricier models turns into a noticeable sum.
Code and development
Despite its positioning, Luna handles code confidently: on the Artificial Analysis Coding Agent Index it beats Claude Opus 4.8. For everyday tasks — write a function, explain someone else's fragment, find a bug in a small file, sketch a script — that is more than enough. A flagship is worth bringing in where long dependent logic or a large codebase is involved.
Long context in the economy class
A 256,000-token window is unusually large for a bottom-tier model. In practice it means a long document or a meeting transcript can be handed over whole, without slicing it up and losing the links between sections. That scenario usually requires stepping up to a pricier model; here it is available on the lowest step.
Vision and files
Luna accepts images in JPG, PNG, and WebP formats, up to 10 files per request. You can attach a screenshot and ask what is on it, pull text from a photo of a document, or have a diagram analyzed. In high-volume scenarios this is handy for processing many images with the same question.
Examples prompt and response
Below are real GPT-5.6 Luna responses produced on Genosai. The first is a content plan, the second explains marketing metrics with a worked calculation. Texts are quoted verbatim, with the long answer shortened.
Prompt: Draft a Telegram post outline for launching a new cosmetics line: 5 bullet points with short subheadings.
GPT-5.6 Luna response:
- **Big Reveal** — Introduce the new cosmetics line and its key promise.
- **What Makes It Special** — Highlight standout ingredients, benefits, or formulation details.
- **Hero Products** — Feature 2–3 must-try products with brief descriptions.
- **Launch Offer** — Share pricing, limited-time discounts, bundles, or free shipping.
- **Shop Now** — Add the launch date, purchase link, and a clear call to action._Generated on Genosai.io with GPT-5.6 Luna._
Prompt: Explain in simple terms the difference between ROAS and ROMI in online advertising, with a numeric example.
GPT-5.6 Luna response:
- **ROAS (Return on Ad Spend)** measures how much revenue your ads generate for each dollar spent on advertising.
- **ROMI (Return on Marketing Investment)** measures the profit generated from the broader marketing investment, taking costs into account.
### Numeric example
Suppose:
- Ad spend: **$1,000**
- Revenue from those ads: **$4,000**
- Other marketing costs—such as creative, software, and agency fees: **$500**
**ROAS:**
4,000 ÷ 1,000 = 4
So the ROAS is **4:1**, meaning you generated **$4 in revenue for every $1 spent on ads**._Generated on Genosai.io with GPT-5.6 Luna._
How to use on Genosai
- Open the model catalog on Genosai and pick GPT-5.6 Luna among the text models.
- Describe the task in the chat: goal, tone, format, and answer length.
- Attach files if needed — a document, data, or an image — and refer to them in the request.
- Send the request and read the answer as it streams, without waiting for the full generation.
- Refine in the same thread: ask to shorten, add an example, or restructure.
With an economy model, precision of wording matters even more. The more concretely you describe the format — how many points, which fields, in what shape — the lower the chance you will have to redo the answer, and the more real the savings become. Genosai runs in the browser with no installation, signing up takes a minute, and billing is by actual usage, so you can start with a single request and judge the model on your own task. Chat history is saved, so work you started is easy to continue later in the same thread.
Prompts
The prompts below play to Luna's strengths: short structured answers, text analysis, and work with large inputs. The general principle is to specify not only the topic but the shape of the answer. The tighter the frame, the more predictable the result.
Classify the sentiment of this review and suggest a short support reply. Format: sentiment, reason, reply.Extract every date, amount, and company name from this text. Return a three-column table.Shorten this product description to 2 sentences, preserving the key specifications and numbers.Split the long document into meaning blocks and give each a 3–5 word heading.Write a JavaScript function that formats a number as a currency amount with thousands separators.Create a 5-step checklist for preparing a webinar announcement post.Analyze the attached screenshot and describe in one paragraph what it shows.Generation cost
On Genosai, GPT-5.6 Luna is billed by tokens — you pay for the actual size of the request and the response, so a short question costs less than a long analysis. The client price is 80 ₽ per 1M input tokens and 480 ₽ per 1M output tokens.
The savings show up not on a single call but at scale. If you have a stream of similar tasks — tagging messages, short descriptions, field extraction — the per-token gap against the Sol flagship is fivefold, and across thousands of calls that becomes a different line in the budget. The sensible approach: keep Luna on the pipeline and bring in pricier models selectively, where quality genuinely decides.
A starter balance after sign-up lets you try GPT-5.6 Luna for free. Current rates and your balance are in the Pricing section.
How it compares
Luna is the bottom step of the GPT-5.6 series, but in quality it does not read like a budget model of the old kind.
| Model | Context (Genosai) | Price ₽/1M (in/out) | Files | Niche |
|---|---|---|---|---|
| GPT-5.6 Luna | 256,000 | 80 / 480 | yes | Economy and volume |
| GPT-5.6 Terra | 256,000 | 200 / 1200 | yes | Balance of power and price |
| GPT-5.6 Sol | 256,000 | 400 / 2400 | yes | Flagship for hard tasks |
| GPT-5.4 Mini | 380,000 | 60 / 360 | yes | Previous generation, cheaper |
For deeper tasks, step up to GPT-5.6 Terra — the price gap is small and the quality headroom noticeable. For genuinely complex agentic scenarios there is GPT-5.6 Sol. From the previous generation, GPT-5.4 Mini comes cheaper but scores lower. Among same-class rivals, compare Claude Haiku 4.5.
A practical rule of thumb: if you are torn between Luna and Terra, look at the shape of your work. Many similar calls — Luna. Complex one-off analyses — Terra. On Genosai both are available in one account, so you can run the same request through two models and compare the answers directly.
Limitations and tips
Luna is the smallest model in the series, and on tasks with long dependent logic it falls behind the larger ones. If an answer comes out shallow or the model drops part of the conditions, that is the signal to move up to Terra rather than to rewrite the prompt for the fifth time.
The higher-effort and parallel-agent modes OpenAI describes are features of its own API. In the Genosai studio the effort level is not exposed and the provider default applies, so plan around the model's ordinary behavior.
A practical tip: on an economy model a strict answer format pays off especially well. Specify the number of points, the column names, the maximum length — that way you get a predictable result on the first try and do not spend extra on retries.
One more technique is to separate the pipeline from the analysis. Send bulk, uniform processing to Luna, and take the individual hard cases it flagged as uncertain to a larger model. That two-stage arrangement is usually cheaper than running the whole volume through an expensive model, and more reliable than trusting all of it to the smallest one.
FAQ
What is GPT-5.6 Luna?
It is the most economical model of OpenAI's GPT-5.6 series, generally available since July 9, 2026. It targets data-heavy, latency-sensitive workloads: chats, classification, lightweight agentic workflows. In the API documentation it corresponds to the nano tier of earlier GPT-5 generations.
How does Luna differ from Terra and Sol?
Sol is the series flagship for the hardest tasks, Terra is the balanced everyday model, and Luna is the most affordable. On Genosai the Luna price per 1M tokens (input/output) is 80/480 ₽ against 200/1200 ₽ for Terra and 400/2400 ₽ for Sol. The per-request gap between Luna and Terra is small, so choose Luna where the volume of calls is what matters.
What is the context window on Genosai?
Up to 256,000 tokens per request, the same as the larger models in the series. For an economy model that is an unusually big window: a long document fits whole, with no slicing into chunks.
How much does a GPT-5.6 Luna request cost?
Billing is by tokens: for the client, 80 ₽ per 1M input tokens and 480 ₽ per 1M output tokens, based on the actual size of the prompt and the response. Per token that is about 2.5 times cheaper than Terra, which makes the model viable for high-volume use.
How capable is Luna for its price?
On the Artificial Analysis Coding Agent Index it outperforms Claude Opus 4.8, at roughly a third of the time and about a quarter of the estimated cost. In knowledge work Luna nearly reaches the peak-performance level of the previous flagship GPT-5.5.
When should you move up to Terra?
Move up to Terra when the task needs deeper reasoning: complex code, long multi-step logic, high-stakes analysis. The price gap between the two is small, so this is not the step to economize on for important work.