OpenAI is tweaking two things at once. Paying subscribers will immediately receive a version of the top-of-the-line model optimized for conversations and a slider for the depth of thought. The free plan will lose its message limit – but not until next week and only for text-only messages.
The GPT 5.6 family has been generally available since July 9th, and OpenAI initially divided it into three tiers: Sol as the top-of-the-line model, Terra for everyday work, and Luna as the fastest and most affordable option. The introduction of the GPT 5.6 models, along with ChatGPT Work, thus marked less of a single new model and more of a new pricing structure. The move on August 6th builds on this and further differentiates the tiers, as OpenAI explains in the announcement for the revised Sol version.
Key Facts at a Glance
- GPT-5.6 Sol is getting a version specifically tailored for conversations, with more direct answers and concise formatting.
- Instant and Thinking will henceforth run on the same model for Plus and Pro, controlled by a slider for the depth of thought.
- Free and Go will switch to GPT-5.6 Luna this week; unlimited text chats and a Think button will follow next week.
- The quotas for file uploads, images and other tools remain unchanged.
- Additional training is intended to ensure answers are available for users under 18 years of age.
A model for Instant and Thinking
The revised version of Sol aims to tailor answers more precisely to the question posed: concise for simple questions, more detailed for multi-stage planning, research, or longer texts. Superfluous formatting is eliminated. Another new feature is that the model will offer disagreement where mere agreement is insufficient.
The more noticeable change is the user interface. Previously, Plus and Pro subscribers chose between Instant and Thinking, each representing a different model with a distinct tone. Going forward, both will run on GPT-5.6 Sol, and a slider determines how much processing power goes into a response. Switching to a higher level should therefore feel like spending more time with the same assistant, not like jumping to a different person.
OpenAI quantifies the gain in reliability based on an internal analysis of questions from finance, medicine, and law. Answers containing at least one factual error occurred approximately 68 percent less frequently than with GPT-5.5 Instant, the previous standard model; for Luna, OpenAI cites 62 percent. Independent verification of these figures is not available.
What each tariff gets – and when
| Tarif | Model in chat | New control system | Available |
|---|---|---|---|
| Free and Go | GPT-5.6 Luna as standard | Think button for more difficult questions | Model changeover this week, unlimited text chats from next week |
| Plus and Pro | GPT-5.6 Sol, revised version | Depth of thought control in web, app and desktop | immediately |
| ChatGPT Work and Codex | GPT-5.6 Sol unchanged | no change | omitted |
The fact that Luna forms the basis for the free tier has a logical explanation. OpenAI reduced the price of this model by 80 percent on July 30th, while Terra saw a 20 percent reduction. The description of the three tiers already positions Luna as the fastest and cheapest option – and what's cheap enough can be offered without limits.
Unlimited applies only to text
The limit is shifting, not disappearing. The quota for text-only chats is being lifted. Existing restrictions remain in place for file uploads, images, and other tools, and OpenAI expressly reserves the right to take action in cases of misuse.
Those who use the free plan for text-based questions, translations, summaries, and code explanations will notice the difference immediately. However, those who regularly upload PDFs or generate images will continue to encounter the same limitations as before.
The Think button is not a shortcut to the top-of-the-line model: it gives Luna more time to respond, but doesn't upgrade her to Sol. The more powerful model remains tied to a subscription.
Stricter limits for users under 18
The update includes additional security training, which OpenAI describes in accompanying system documentation. For accounts that the company assigns to younger users, the model is designed to block romantic role-playing, not to be seen as a substitute for real-life relationships, and to reject age-restricted dares.
This includes limits on sexual content, eating disorders and body image, age-restricted goods, dangerous activities, and graphic violence. Where a young person needs support, the model aims to encourage them to talk to trusted adults. OpenAI supplements the training with system-level protection mechanisms and its own testing procedures for this age group.
How much of that actually reaches Germany, Austria and Switzerland
OpenAI has not specified any country restrictions for this update. This is noteworthy because this has been rather the exception for ChatGPT in recent months. The ChatGPT release notes reveal a number of features that were initially excluded from the European Economic Area, the UK, and Switzerland: computer usage under Windows in Codex, the Record & Replay function, the public publishing of ChatGPT sites, and the ChatGPT Health service.
It's important to distinguish between Switzerland and the EU: Switzerland is neither in the EU nor the EEA. The fact that OpenAI regularly excludes it along with the EEA is a company decision and not a consequence of European regulation. Apple's approach is the opposite – features that are unavailable in the EU are generally available to Swiss users.
A second regional point concerns advertising. OpenAI initially launched it in the US for the free and Go plans, and subsequently expanded it to Canada, Australia, New Zealand, and the UK. Germany, Austria, and Switzerland are not mentioned in the previous announcements regarding the introduction of advertising in ChatGPT. Therefore, the free plan will be unlimited in these countries, without any ads interfering.
It remains unclear what the update means for ChatGPT integration within Apple Intelligence. OpenAI explicitly mentions the chat environment in the ChatGPT apps for web, mobile devices, and desktop. The announcement makes no mention of the extension that can be configured on the iPhone within Apple Intelligence.
Why "newest models" isn't called "most powerful models"
Unlimited text chats sound like a bigger concession, but they're also the cheaper option. Luna is the most affordable tier, and the price reduction at the end of July has made it so much more affordable that a limit is hardly necessary from a business perspective. The top-tier model remains behind a paywall, and with the merging of Instant and Thinking, it's even closer to the everyday lives of subscribers than before.
The scale at which this is happening is remarkable: OpenAI now estimates usage at one billion people per week. At the beginning of the year, ChatGPT was only just approaching this mark – dropping a limit on the number of conversations would not have been feasible without the efficiency gains of recent weeks.
For you, this means that those using ChatGPT without a subscription primarily gain in volume. Extensive research, complex coding tasks, and multi-stage planning remain the domain of paid tiers. The gap between the plans widens with this change; it simply shifts from the question of "how often" to "how well." Even those previously held back by message limits now benefit more, as a depleted quota has resulted in more conversations being concluded than a missing model tier. (Image: OpenAI / Apfelpatient)
- Apple Vision Pro reduces surgery time by 19 percent
- Apple's leverage against memory prices is fizzling out
- OpenAI requests that Apple's lawsuit be dismissed in court
