GPT-6 Luna Explained: OpenAI’s Incredible Budget AI Model

GPT-6 Luna is the most affordable model OpenAI has ever released — and since early October 2026, it has also become the model that most free ChatGPT users actually talk to. It costs a hundredth of what the flagship GPT-6 Astra charges per token, it answers in plain conversational language, and OpenAI is now rolling it out to the free tier with a brand-new interactive interface that builds charts, calculators and forms inside the chat itself.

If you have noticed a faster, more helpful free ChatGPT in the past few days, that is almost certainly GPT-6 Luna at work. This guide explains what it is, what it costs (in real US dollars), what it can and cannot do, and how it stacks up against rivals like Claude Haiku 5.5 and its own siblings, GPT-6 Sol and GPT-6 Astra. Every price and specification below comes from OpenAI’s published pricing and reporting from early October 2026.

Woman chatting with an AI assistant on a laptop, like GPT-6 Luna

What Is GPT-6 Luna?

GPT-6 Luna is the smallest, cheapest member of OpenAI’s GPT-6 family. OpenAI launched it on 22 September 2026 alongside GPT-6 Sol, framing the pair as the way “the benefits of that intelligence” — referring to the flagship GPT-6 Astra — get distributed more widely. In plain terms: Astra showed what the new generation could do, and Luna takes a lighter-weight version of that training and sells it at a fraction of the price.

Think of the GPT-6 family as a ladder with three main rungs:

  • GPT-6 Astra — the flagship. The strongest model, the most expensive, the one used for the hardest reasoning and coding work.
  • GPT-6 Sol — the mid-tier. Aimed at everyday professional work: coding, documents, longer analysis. Later refined into GPT-6.1 Sol.
  • GPT-6 Luna — the budget tier. Built for high-volume, fast, routine jobs: summarising documents, extracting information, answering quick questions.

Luna’s design philosophy is the opposite of a showcase model. It is optimised for speed and volume: it responds quickly, it handles an enormous context window, and it is cheap enough that developers can call it thousands of times a day without flinching. If Astra is the specialist you consult for the hard cases, Luna is the tireless assistant you hand the everyday pile to.

The model is identified as gpt-6-luna in OpenAI’s API, and it became available to developers at the same time as Sol. But for ordinary users, the real news came two weeks later: OpenAI began rolling Luna out to free and Go-tier ChatGPT accounts from 8 October 2026, which is why searches for the model have spiked this month.

Three things have pushed Luna into the spotlight in the last few weeks:

1. Free users finally get a GPT-6 model

Starting 8 October 2026, OpenAI began giving free and Go-tier ChatGPT users access to GPT-6 Luna, with Plus, Pro, Business and Enterprise users receiving GPT-6 Sol in the chat interface. The rollout is staged — not every account gets it on day one — but the direction is clear: the free tier of ChatGPT is now powered by genuinely current-generation AI, not a stale older model. For the more than a billion people who use ChatGPT weekly, this is the first time a model from the GPT-6 family has been available for nothing.

2. Intelligent UI turns answers into interactive tools

Alongside the free rollout, OpenAI introduced what it calls Intelligent UI: the model automatically generates interactive elements — diagrams, charts, calculators, forms, maps and timers — directly inside the conversation, without the user having to ask for them. Ask for a mortgage comparison and Luna draws the chart; ask for a recipe schedule and it builds the timer. Reporting on the launch notes that OpenAI’s internal evaluation found the new generation responds to web-search questions about 44% faster than its predecessor, using a technique called interleaved thinking that delivers an initial answer while continuing to work in the background. Whether those vendor figures hold up in independent testing remains to be seen, but the experience itself is the draw: chat that builds tools on the fly.

Our full GPT-6 Intelligent UI guide walks through exactly what the interface can generate and where it still struggles.

3. The price war with Anthropic is heating up

OpenAI cut Luna’s API prices in half on launch day — and Anthropic answered the same week by launching Claude Haiku 5.5 at exactly Luna’s price ($0.10/$0.50 per million tokens) while claiming better benchmark scores. That head-to-head battle between two ultra-cheap frontier-lite models is exactly what developers and finance teams are watching right now, because whichever wins will power a huge share of the world’s AI agents in 2027.

GPT-6 Luna Pricing: The $0.10 Model Explained

Pricing is where Luna genuinely breaks from the pack. Here is what OpenAI charges, per one million tokens, based on the published pricing tables:

Pricing tier Input (per 1M tokens) Output (per 1M tokens)
Standard $0.10 $0.50
Cached input (repeat context) $0.01 —
Batch mode (non-urgent) $0.05 $0.25
Very long prompts (272K+ tokens) $0.20 $0.75

For context: these launch prices are half of what the previous generation cost. GPT-5.6 Luna charged $0.20/$1.20, so the new model halves the bill while being a newer, better-trained model. The standard rate is also exactly 100 times cheaper than GPT-6 Astra’s $10/$50 rates — the steepest discount OpenAI has ever put between a flagship and its budget sibling.

A few details worth knowing before you budget:

  • Cached input reads cost almost nothing. At $0.01 per million tokens — a 90% discount on the standard input rate — Luna rewards designs where the model re-reads the same context (for example, an agent working through a long document across many steps). Agents that reuse context heavily get the cheapest rates of all.
  • Batch mode halves the price again. If your work is not time-sensitive, batch requests run at $0.05 input and $0.25 output per million tokens. For bulk classification, summarisation or data extraction jobs, that is genuinely remarkable: a million tokens of processing for a nickel.
  • The very-long-prompt tier exists but is generous. Only prompts above roughly 272,000 tokens trigger the higher $0.20/$0.75 rates. Few everyday tasks get anywhere near that threshold, and even then the price is modest.
  • Web search costs extra. OpenAI charges about $10 per 1,000 web-search calls when the model uses its search tool. Factor that in if your application searches the web constantly.
  • Cache writes have their own line item. Writing to the prompt cache is priced at $0.125 per million tokens — worth knowing if you rebuild large cached contexts frequently.

What does this mean in everyday terms? A one-million-token job — roughly 750,000 English words, or about ten full novels — costs ten cents on the input side. If you are a student summarising lecture notes, a marketer classifying reviews, or a developer building a high-call-volume agent, GPT-6 Luna is priced so low that the model cost barely registers on the bill.

GPT-6 Luna pricing vs the rest of the GPT-6 family

Model Input / 1M Output / 1M Best for
GPT-6 Luna $0.10 $0.50 Fast, high-volume routine work
GPT-6 Sol / 6.1 Sol $2.00 $10.00 Coding, agents, professional work
GPT-6 Astra $10.00 $50.00 The hardest reasoning tasks

The pricing story is simple: Sol costs 20x what Luna costs, and Astra costs 100x. OpenAI clearly expects the overwhelming majority of high-volume traffic — agents, automation, everyday chat — to run on Luna, with the expensive tiers reserved for work that truly needs them. Our GPT-6.1 Sol guide covers the mid-tier in depth if you want the comparison.

Person typing into a ChatGPT-style chat window, the interface GPT-6 Luna now powers for free users

What GPT-6 Luna Can Actually Do

Cheap does not mean dumb. Here is a realistic picture of where Luna performs well — and where it does not.

Intelligent UI: answers that build things

The headline consumer feature is Intelligent UI. Instead of replying with text alone, GPT-6 Luna decides on its own when a visual or interactive element would help, and generates it on the spot: a chart for data, a calculator for numbers, a form for structured input, a map for locations, a timer for scheduling. This works for free-tier users too, which makes it one of the biggest free-tier upgrades ChatGPT has ever received. One caveat from launch coverage: older Windows and macOS desktop apps do not support the new interface — if you are stuck on an old app version, use the web version instead.

Coding and agentic work

Despite its budget billing, GPT-6 Luna carries a serious context window — about 1.05 million tokens, with up to 922,000 tokens allowed in a single input and up to 128,000 tokens of output. It also offers six reasoning-effort settings, a dial developers can turn between faster answers and deeper thinking. On OpenAI’s own DeepSWE 1.1 coding benchmark, Luna scored 66.6% — behind Sol’s 68.8%, but in the same league.

Be realistic about the trade-offs, though. Independent testing after launch found mixed results: small regressions on some coding evaluations compared with expectations, even though OpenAI’s own figures showed improvements over the previous generation. For simple agents, data extraction and routine coding assistance, Luna is excellent value; for the hardest engineering work, the mid-tier models remain the safer bet.

A calmer, clearer communication style

OpenAI says the GPT-6 generation introduced an improved communication style that Sol and Luna both inherited: simpler conversations, less jargon, fewer low-value details, and — in OpenAI’s own alignment testing — lower rates of misleading claims about the model’s own coding work. In practice, answers from Luna feel noticeably more concise than the older models’ habit of burying the answer under paragraphs of filler.

What GPT-6 Luna does NOT do

No GPT-6 model currently generates images, video or audio — Luna included. It is a text (and tool-use) model. If your project needs generated visuals or narration, you will need a separate model for that part of the pipeline. Also, Luna is classified by OpenAI as High capability in cybersecurity and biological/chemical domains — a step below Astra’s Critical rating — which reflects its lighter-weight training.

GPT-6 Luna vs Sol, Haiku 5.5 and Astra

Here is how GPT-6 Luna compares with the models buyers are actually cross-shopping it against:

GPT-6 Luna GPT-6 Sol / 6.1 Sol Claude Haiku 5.5 GPT-6 Astra
Launched 22 Sep 2026 22 Sep / 29 Sep 2026 7 Oct 2026 3 Sep 2026
Price (in/out, per 1M) $0.10 / $0.50 $2.00 / $10.00 $0.10 / $0.50 $10.00 / $50.00
Context window ~1.05M tokens ~1M+ tokens 1M tokens ~1M+ tokens
DeepSWE 1.1 (vendor) 66.6% 68.8% (Sol) Not directly comparable 77.9%*
Free ChatGPT access Yes (from 8 Oct) No (paid tiers) Via Claude app/API Paid tiers only
Best for Volume, speed, cost Coding, agents, pro work Computer use, terminals Hardest reasoning

*Astra’s figure is Google’s reported Gemini 4 Argon benchmark context; vendor-reported numbers are not directly comparable across companies. See our Gemini 4 Argon guide for how the frontier race looks from the other side.

The GPT-6 Luna vs Claude Haiku 5.5 fight is the one to watch. Anthropic priced Haiku 5.5 at exactly Luna’s rates and claims it beats Luna on every benchmark the two share — including big gaps on desktop operation and terminal tasks in Anthropic’s own comparison. The honest caveat: Anthropic has not said whether it ran Luna itself or used OpenAI’s published figures, so treat that comparison as vendor-reported until independent testers weigh in. What is not in dispute is that both models are absurdly cheap, both have million-token-class context, and both are gunning for the same high-volume agent workloads.

Freelancer smiling while working on a laptop, everyday work GPT-6 Luna is built for

Who Should Use GPT-6 Luna — And Who Should Skip It

GPT-6 Luna is a great fit if you:

  • Run high-volume automation. Customer support triage, review classification, document summarisation, data extraction — anywhere you make thousands of model calls a day, Luna’s ten-cent pricing turns AI from a budget line item into a rounding error.
  • Build agents. The huge context window, cheap cached reads and six reasoning-effort settings make Luna a natural default brain for routine agent steps, with pricier models reserved for the hard ones.
  • Just want a better free ChatGPT. If you are on the free tier, the October rollout means you are probably already using Luna. The Intelligent UI alone — charts, calculators, forms generated inside the chat — makes it worth checking what changed.
  • Prototype cheaply. Testing an AI feature idea? Luna lets you iterate through thousands of test calls for the price of a coffee.

Hold off on GPT-6 Luna if you:

  • Need maximum coding performance. Independent testing showed mixed results on harder coding evals. GPT-6.1 Sol or a larger model is the safer choice for complex engineering.
  • Need generated images, video or audio. Luna does none of these. Budget a multimodal model alongside it.
  • Depend on deterministic reasoning effort. The six reasoning-effort settings are a nice dial, but heavy agentic pipelines may prefer the deeper reasoning of Sol-class models for critical-path steps.
  • Live at the very-long-prompt extreme. If your prompts routinely exceed ~272,000 tokens, the doubled rates and Haiku 5.5’s 100K-token cliff both deserve a careful spreadsheet comparison first.

How to Start Using GPT-6 Luna Today

Getting started depends on who you are:

  1. Free ChatGPT users: open the ChatGPT web app (or update to the latest desktop/mobile app — older versions do not support Intelligent UI) and just start chatting. The October 2026 rollout is staged, so if your account has not received it yet, it should arrive shortly. Ask it something that benefits from a visual — “compare these two phone plans” or “plan my week” — and watch what it builds.
  2. Paid ChatGPT users (Plus, Pro, Business, Enterprise): you get GPT-6 Sol in the chat interface; Luna remains available through the API and in ChatGPT Work and Codex for building.
  3. Developers: call the gpt-6-luna model ID through OpenAI’s API. Start with the standard tier; if your workload reuses long contexts, switch on prompt caching for the $0.01/1M cached-input rate; if the work is not urgent, route it through batch mode at $0.05/$0.25.
  4. Budgeting tip: instrument your token usage by call type for a week. Teams are consistently surprised by how much of their spend is re-readable context — which is exactly where Luna’s cached-input discount does its best work.

GPT-6 Luna FAQ

Is GPT-6 Luna free?

The model itself is an API product priced at $0.10/$0.50 per million tokens, but OpenAI has been rolling it out to free and Go-tier ChatGPT users since 8 October 2026. If you use the free ChatGPT app or website, you are very likely already chatting with Luna at no cost.

Is GPT-6 Luna better than Claude Haiku 5.5?

It depends on the task — and on whose numbers you believe. Anthropic’s own launch comparison says Haiku 5.5 beats Luna on every shared benchmark, but Anthropic has not disclosed whether it ran Luna itself. Independent testers have found Luna’s coding scores mixed. Both cost exactly the same, so the honest answer is: test your specific workload on both before committing.

What is GPT-6 Luna’s context window?

About 1.05 million tokens, with up to 922,000 tokens accepted in a single input and up to 128,000 tokens of output. Prompts above roughly 272,000 tokens are billed at higher rates — precisely $0.20 input and $0.75 output per million tokens.

Does GPT-6 Luna make images or video?

No. Like every GPT-6 model released so far, Luna is a text and tool-use model. It cannot generate images, video or audio. It can, however, generate interactive elements like charts, calculators, forms and timers inside the chat through Intelligent UI.

How much cheaper is GPT-6 Luna than GPT-6 Astra?

Exactly 100 times cheaper on both input and output: $0.10/$0.50 versus Astra’s $10/$50 per million tokens. It is also half the price of its own predecessor, GPT-5.6 Luna, which charged $0.20/$1.20.

Person using an AI assistant on a smartphone with a laptop nearby, a typical GPT-6 Luna use case

Final Verdict

GPT-6 Luna is the clearest sign yet of where the AI industry is heading: frontier-class training, priced like a utility. It will not beat GPT-6 Astra or GPT-6.1 Sol on the hardest tasks, and the jury is still out on whether it or Claude Haiku 5.5 deserves your high-volume workloads. But for everyday work — the summarising, the classifying, the quick questions, the thousand small calls an agent makes per hour — Luna’s combination of a million-token context, Intelligent UI and ten-cent pricing is genuinely hard to argue with.

For most readers, the takeaway is simple: open the free ChatGPT app, try the new interactive answers, and notice how much of your daily AI use fits comfortably inside a model that costs almost nothing to run. The expensive models still have their place — but after Luna, that place is a lot smaller than it used to be.

All benchmark figures in this article are vendor-reported unless stated otherwise, and prices reflect OpenAI’s published API pricing as of early October 2026. Model rollouts are staged, so free-tier availability may vary by account. For the broader frontier-model picture, read our GPT-6 Intelligent UI guide and our Gemini 4 Argon explained article.

Leave a Comment

Your email address will not be published. Required fields are marked *