Is GPT-6 Luna Free

Yes. GPT-6 Luna is free to try for Free and Go users in the desktop app, but that does not make it free everywhere: ordinary Chat did not have it at launch, Work and Codex access sits on paid plans, and the API starts at $0.10 per million input tokens plus $0.50 per million output tokens.

The useful answer is not a single price. It is a map of plan, product surface, and billing method. Pick the wrong surface and “free” turns into either a missing model selector or a separate API bill.

Is GPT-6 Luna Free? The Short Answer

GPT-6 Luna is free on one clearly documented route: sign in to the desktop app with a Free or Go account. OpenAI did not put the model in ordinary Chat at launch, and it did not attach free API credits to that desktop entitlement.

That matrix comes from OpenAI's September 22 launch announcement. The important GPT-6 Luna ChatGPT distinction is between ChatGPT as the account and Chat as a product surface. A Free account can have desktop-app access while the Chat surface still lacks the model.

What Actually Changed on September 22

GPT-6 Luna replaced an older free-tier story with a new surface-specific one. OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22, 2026, then made Luna available to Free and Go accounts in the desktop app while reserving Work and Codex access for paid plans.

That is not the same rollout announced in August. GPT-5.6 Luna had become the Free and Go default for text chats. GPT-6 Luna arrived as a new model with a different initial access boundary, so the older “unlimited text chats” language cannot be copied forward.

Capability improved as price fell, according to OpenAI's own evaluations. The company reports that GPT-6 Luna at high effort gained 5.4 percentage points on AutomationBench while costing 58% less per task than its predecessor. Treat that as vendor evidence, not as proof that a free desktop account will have the throughput, tools, or limits your workflow needs.

GPT-6 Luna Free Access: What OpenAI Includes

The free entitlement includes the model in the desktop app, not a universal pass across every OpenAI product. Free and Go are the two named eligible account types. Plus, Pro, Business, Enterprise, and Edu are the named paid plans for Work and Codex.

OpenAI also said the launch-day rollout would happen gradually. If the model was missing during the day, its advice was to try again later. A missing selector on September 22 therefore could mean rollout timing rather than account ineligibility.

This access is documented, not tested here. The desk supplied no Free or Go account, so no claim is being made about the current desktop app version, selector placement, task quality, or a limit message seen in use. The existing ChatGPT Free vs Plus comparison still covers the older general Chat experience; the new GPT-6 Luna decision turns on its narrower surface boundary.

GPT-6 Luna Desktop: How to Verify Your Entitlement

The cleanest check is a small, recorded evaluation that separates account access from model quality. Do not start with client data or a production repository. Use a synthetic task you can repeat elsewhere.

1. 1

   Record the account

   Write down whether the signed-in account is Free or Go, the desktop app version, and the date. This makes the observation useful after the rollout changes.

2. 2

   Capture the selector

   Open the model selector. If GPT-6 Luna appears, record its exact label and take a screenshot. If it does not, record that too instead of assuming the account is ineligible.

3. 3

   Run one synthetic task

   Use a small work sample with no sensitive data, such as summarizing a fictional support queue or revising a short mock product brief. Save the prompt and output.

4. 4

   Compare ordinary Chat

   Open the Chat selector with the same account and record whether GPT-6 Luna appears there. At launch, OpenAI documented that it was not yet available in Chat.

5. 5

   Save the limit message

   If the app shows a usage warning, cooldown, or reset time, capture the exact wording. Do not convert one account's observed limit into a promise for every plan.

This protocol produces evidence you can act on without pretending that a launch announcement describes every account state.

What It Means for Builders, Operators, and Buyers

The launch changes the evaluation cost first, not every production decision.

Builders can validate before opening API billing

A solo technical builder can use the free desktop route to see whether GPT-6 Luna follows instructions, handles a representative code explanation, and communicates at the required level. If it fails the work sample, API pricing is irrelevant. If it passes, move the same synthetic task to the API and measure tokens before connecting a real application.

The wall appears when you need automation. The launch announcement does not turn desktop entitlement into an API key or document production latency guarantees or free token usage. Programmatic calls use a separate billing system.

Operators should treat free access as a trial lane

A senior operator can evaluate a recurring work task without asking procurement for another seat. That is useful for intake triage, document comparison, or drafting an internal plan. It is not enough to make the desktop app a dependency for a weekly process because the launch announcement gives no numeric usage ceiling.

The decision rule is simple: evaluate now, standardize later. Require a stable access path, an observed limit, and an owner for failures before the task becomes operational.

Buyers should purchase the surface, not the model name

A buyer who only wants to try Luna should not upgrade merely because GPT-6 is new. A buyer who needs Work or Codex has a paid-plan reason. A team building an application has an API reason, which is a separate budget from the ChatGPT subscription.

That distinction also keeps the ChatGPT pricing decision clean: a plan buys access to ChatGPT surfaces, while the API buys metered programmatic usage.

GPT-6 Luna API Cost: The Numbers

GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens at Standard short-context rates. The live OpenAI API pricing page also lists $0.01 per million cached input tokens and $0.125 per million cache-write tokens for that tier.

The practical comparison is a workload, not one token price. For a 5:1 mix of input to output, the new rates cut the following bills:

The first row is the useful unit: $0.10 of input plus $0.10 of output. The predecessor cost $0.20 of input plus $0.24 of output for the same mix. OpenAI describes the family price cut as 50%; the workload-specific reduction is 54.5% because the output rate fell from $1.20 to $0.50.

These totals can be spread across many short-context requests. They do not claim that one request can consume the whole monthly workload at short-context rates.

The API is not bundled with a ChatGPT subscription. OpenAI's billing guidance says the two products use separate billing systems and that API usage is billed separately.

GPT-6 Luna Codex Access Is a Paid-Plan Benefit

Codex is not part of the free desktop promise. OpenAI named Plus, Pro, Business, Enterprise, and Edu as the plans with GPT-6 Luna access in Codex and ChatGPT Work at launch.

For a developer, this creates three distinct tests. Desktop answers whether the model is worth exploring. Codex answers whether it fits repository work inside a paid product. The API answers whether it fits an application that your software calls directly. Success in one surface does not prove identical tools, system instructions, or limits in another.

Upgrade only if the surface matters. Paying merely to obtain the Luna name is unnecessary when the desktop evaluation route is already available to Free and Go users.

GPT-6 Luna Limits: What OpenAI Has Not Published

The launch announcement does not publish a numeric desktop cap for GPT-6 Luna. It does not specify messages per day, tasks per week, a cooldown duration, or a reset schedule for Free and Go accounts.

That omission matters because the previous generation used different language. OpenAI's August GPT-5.6 announcement described unlimited text chats for GPT-5.6 Luna. The September GPT-6 announcement documents desktop access but does not repeat that promise. An older unlimited claim is not a current GPT-6 limit.

Plan around what is known:

• Free and Go desktop entitlement is documented.
• A specific GPT-6 Luna desktop allowance is not documented in the launch post.
• Work, Codex, Chat, and API are separate surfaces with different access or billing rules.
• A limit observed on one account should be recorded with its plan, app version, and date.

For a production dependency, wait until your own account evidence or updated official documentation gives you a stable ceiling.

What Is Overhyped About “Free”

“Free” is accurate only when the sentence ends with in the desktop app for Free and Go users.

Three broader interpretations are not supported:

1. “GPT-6 Luna is free in every ChatGPT surface.” It was not yet available in ordinary Chat at launch.
2. “Free access is confirmed unlimited.” OpenAI published no numeric GPT-6 Luna desktop cap and did not carry the GPT-5.6 unlimited-text wording into this launch.
3. “A ChatGPT plan includes API usage.” API billing is separate and token-metered.

The vendor benchmarks can also be oversold. A 66.6% DeepSWE v1.1 result at max effort is a useful capability signal, but it does not tell you whether your free account has enough usage for a workday or whether your task will succeed. Access, allowance, and ability are three different questions.

How to Use GPT-6 Luna This Week

Act now if you already have a Free or Go account and can use the desktop app. Run one synthetic task, save the exact model label and any limit notice, and compare the output with the model currently used for that task.

Act now on the API if you are a builder with a bounded pilot. Set a spend limit, send the same work sample, and calculate the cost from actual input and output tokens. Do not connect an open-ended retry loop on day one.

Wait if your workflow requires ordinary Chat, a published free-tier quota, or consistent cross-account availability. Those conditions were not established by the launch announcement.

You are largely unaffected if GPT-5.6 in ordinary Chat already meets your needs and you have no use for Work, Codex, or an API integration. A new model name is not a reason to move a stable workflow.

The Monday move: record your plan and desktop app version, run one low-risk work sample in GPT-6 Luna, capture any limit shown, and decide separately whether Work, Codex, or the API earns a budget.

Frequently Asked Questions

Why is GPT-5.6 Luna so cheap?

GPT-5.6 Luna was designed as the cost-efficient model in its generation. GPT-6 Luna pushes that position further with lower input and output prices, but “cheap” describes API economics, not a guarantee of unlimited free use.

How much does GPT-5.6 Luna cost?

OpenAI's GPT-6 launch comparison lists GPT-5.6 Luna at $0.20 per million input tokens and $1.20 per million output tokens. Current GPT-6 Luna Standard short-context pricing is $0.10 input and $0.50 output.

What is GPT-5.6 Luna good for?

OpenAI made it the Free and Go default for text chats in August 2026, so everyday text was its clearest documented role. GPT-6 Luna is now current; use the older model only when a provider or existing workflow specifically depends on it.

Is GPT Luna any good?

OpenAI reports a 66.6% DeepSWE v1.1 score for GPT-6 Luna at max effort, but this desk did not test the model. The honest answer comes from a representative task plus the limit and cost your account shows.

Which is better, GPT-5.6 Luna or Terra?

That is a previous-generation choice. The GPT-5.6 family review covers Sol, Terra, and Luna; it should not be used to infer GPT-6 Luna's current entitlement or limits.

Which is better, GPT-5.6 Luna or GPT-5.3 Codex?

Neither is universally better: Luna targeted low-cost general and coding work, while a Codex model is selected for agentic coding. For the current launch, the actionable fact is that GPT-6 Luna itself is available in Codex on paid plans.

What are the differences between GPT-5.6 Luna and Luna Pro?

The September 22 materials do not define a GPT-6 model called Luna Pro. Do not map an older provider label or informal name onto GPT-6 Luna plan access without a current vendor page.

Which GPT Codex is the best?

The best Codex model depends on repository difficulty, latency, and budget. GPT-6 Luna is the low-cost new option named in this launch, but a controlled task comparison is more useful than a universal ranking.

Is 5.6 Luna good for coding?

This run did not re-test GPT-5.6 Luna or lock a current independent coding score for it. OpenAI reports coding gains for GPT-6 Luna, but those are vendor results, so validate merge quality and instruction following on your own synthetic repository task.

For dated, practical notes when AI access and pricing change, join the newsletter.