Free and Go now default to Luna

OpenAI is rolling out GPT-5.6 Luna as the default model for ChatGPT Free and Go. Luna is the fastest and most affordable member of the GPT-5.6 family, which also includes Terra and flagship model Sol.

The most visible change is expanded text conversation without a rate limit, subject to abuse guardrails. For ordinary use involving questions, drafting, summarization or conversational help without additional tools, the message counter becomes much less central.

The important limitation is that OpenAI still applies separate caps to file uploads, images and other tools.

Think is not a free button for GPT-5.6 Sol

Free and Go users also receive a Think button for harder questions. Technically, it gives GPT-5.6 Luna more time to reason through the request.

OpenAI's current documentation explicitly says Think does not use GPT-5.6 Sol. Free and Go users do not have access to Sol in standard ChatGPT conversations.

That distinction is essential to understanding the product. The button allows the same model tier to spend more effort on a request; it does not substitute for the higher capability tier offered on paid plans.

Sol remains tied to higher reasoning levels

On eligible plans, GPT-5.6 Sol powers Medium and High reasoning options. Higher subscriptions can also receive additional effort levels, including GPT-5.6 Sol Pro for the most demanding work.

OpenAI also provides a reasoning control for eligible paid users. A person can favor a faster response or allocate more thought to programming, research, planning, writing or another complex task.

That makes ChatGPT more complicated than a simple choice between a free model and a paid model. Three dimensions now overlap: model tier, reasoning effort and tool access.

Luna is not presented as a throwaway model

Luna sits at the low end of GPT-5.6 in cost and latency, but OpenAI does not position it as a basic fallback. The company says it retains substantial capability from the 5.6 generation while targeting fast and extremely high-volume workloads.

In an internal evaluation involving financial, medical and legal questions requiring factual detail, OpenAI says responses containing at least one factual error were around 62 percent less common with GPT-5.6 Luna than with GPT-5.5 Instant.

That number needs context. It comes from an OpenAI internal evaluation on a specific set of prompts and is not a guarantee that every Luna conversation will be 62 percent more accurate.

Lower cost is part of the product design

Serving a much larger volume of conversation is not purely a commercial decision. OpenAI has presented the entire GPT-5.6 family around efficiency, with different models designed to trade capability, speed and cost against one another.

OpenAI also cut Luna's API price by 80 percent on July 30. Its public rate is now 20 cents per million input tokens and 1.20 dollars per million output tokens.

Those API prices do not reveal what a ChatGPT conversation actually costs OpenAI. They do show that Luna has been engineered and marketed as the very low-cost layer of the GPT-5.6 generation.

Unlimited really means limits are becoming more targeted

For a free user, the everyday consequence is positive and straightforward: a text conversation can continue much longer without running into a conventional message quota.

Expensive resources have not become free, however. Images, files and tools retain their own caps, while Sol reasoning remains restricted to plans that include it.

The important change is therefore not the disappearance of limits. It is that OpenAI is beginning to align those limits more closely with the actual cost and complexity of different actions rather than counting every message in the same way.

For users, that is probably a better abstraction: conversation becomes the baseline, while the heavier operations are where ChatGPT reminds you that large-scale AI is still a computing service with finite resources.