Starting this week, ChatGPT free users can send text messages without hitting a usage cap. OpenAI is also upgrading the default model for Free and Go accounts from GPT-5.5 Instant to GPT-5.6 Luna, and rolling out a “Think” button for on-demand reasoning. Three changes in one announcement, and the most consequential for most users is the first one.
What exactly is changing for free ChatGPT users?
Until now, free users who sent too many messages in a short window would hit a rate limit (a cap on how frequently requests are accepted). The result: wait, or drop to a more restricted experience. According to OpenAI’s announcement covered by Dân trí on August 7, 2026, that limit is being removed for text-only conversations. File uploads, image generation, and other tools still have usage caps.
At the same time, the default model for Free and Go accounts is being upgraded from GPT-5.5 Instant to GPT-5.6 Luna. In OpenAI’s internal evaluation covering finance, health, and legal questions requiring high factual accuracy, GPT-5.6 Luna produced roughly 62% fewer responses containing at least one factual error compared to GPT-5.5 Instant, according to the same report.
Free users are also gaining a “Think” button. When activated, it triggers extended reasoning mode, meaning the model spends more time working through a problem step by step before responding. This is opt-in: users choose when to turn it on, rather than having it run on every query.
What are Plus and Pro users getting in this update?
Paid subscribers are getting GPT-5.6 Sol, a model tuned for more direct responses. The core change is adaptive answer length: simple questions get short answers, while complex tasks like research, multi-step planning, or coding get longer responses that OpenAI says stay on topic and avoid unnecessary formatting. In the same internal evaluation, GPT-5.6 Sol showed roughly 68% fewer responses with at least one factual error compared to GPT-5.5 Instant, in finance, health, and legal test categories, per OpenAI’s announcement.
Plus and Pro users also get a “thinking slider” instead of a binary button. The slider lets users dial how much reasoning time the model allocates per query. Lower settings favor speed; higher settings favor depth for research, planning, writing, coding, or complex decisions.
Why is OpenAI making these changes now?
OpenAI framed the free-tier expansion as helping more people ask questions, develop ideas, and get support without interruption. Some observers quoted in the Dân trí report from August 7, 2026 see the timing differently: Gemini and other AI assistants have been actively expanding their own free-tier offerings. Removing friction for free users is a direct way to compete for that base. Both readings can be true at once.
What the upgrade does not change: non-text usage. File upload limits, image tools, and other features beyond text remain capped for free users. The no-limit policy applies specifically to text conversation.
What changed across tiers at a glance
- Free and Go: No more text message rate limit; model upgraded from GPT-5.5 Instant to GPT-5.6 Luna; “Think” button added for on-demand reasoning.
- Plus and Pro: Model upgraded to GPT-5.6 Sol; new thinking slider for adjustable reasoning depth per query.
- All tiers: File upload, image, and tool usage caps remain in place for free accounts.
What this means for AI-search visibility
Here is the practical upshot for any brand that wants to appear in AI-generated answers: more users querying ChatGPT without interruption, using a model that makes fewer factual mistakes, means more AI-mediated searches happening at scale every day.
Think of it this way. Before this update, a free user hitting the rate limit would stop, switch tabs, maybe open a browser and do a regular Google search instead. That is a moment your content could have been surfaced by AI, lost. Remove the cap, and that user stays inside ChatGPT. The query happens. The model answers. Your content is either cited or it is not.
The 62% and 68% factual-error reductions for Luna and Sol respectively matter here for a specific reason the source does not spell out: a more accurate model is a more confident model. When AI tools are uncertain about a fact, they tend to hedge or omit citations. When they are more confident, they are more likely to attribute a specific answer to a specific source. Brands with well-structured, fact-dense, clearly attributed content have more surface area to be cited in this environment.
The Think button and thinking slider add another layer worth watching. When a user activates extended reasoning on a complex question, the model is likely drawing on sources it considers more authoritative or more specifically relevant than in fast mode. This is not confirmed by OpenAI’s announcement, but it is a reasonable inference from how extended reasoning architectures (systems designed to work through multi-step problems) generally operate. The exact retrieval behavior depends on the platform’s implementation and will vary.
Hingewise’s read: the removal of the text message cap matters more for AI-search visibility than the model upgrade does. A free-tier user who now stays in ChatGPT for a longer, uninterrupted session generates more queries, and each of those queries is an opportunity for content to be found, cited, or recommended. The model accuracy improvements make individual citations more reliable, but it is the volume shift, driven by removing the cap, that changes the scale of the opportunity. Brands that have not yet audited how their content appears (or fails to appear) in AI-generated answers now have less time before competitors who have done that work pull ahead in Share of Model (the proportion of AI responses in a given topic where your brand gets mentioned vs. competitors).
Five things to check before assuming your content is AI-ready
- Can ChatGPT’s search and browsing features actually find your key pages? (Check if your robots.txt or other access controls block AI crawlers.)
- Does your content answer specific, factual questions in the categories where AI accuracy is improving fastest: finance, health, legal, and similar high-stakes topics?
- Are key facts in your content attributed with sources and dates, the way a model trained on factual reliability would prefer to cite?
- Does your core content hold up as text only, given that non-text tools remain capped for free users and text is what the model now handles at unlimited scale?
- Have you measured your current Share of Model for your main topic areas? If not, you have no baseline to know whether this update helps or hurts your AI visibility.
The rollout begins this week. The question worth tracking over the next few months is whether removing the free-tier cap produces a measurable shift in the volume and nature of queries reaching AI search systems, and whether the factual accuracy improvements in Luna and Sol translate to more specific, source-attributed answers rather than hedged generalities. That is where AI-search visibility either grows or stalls.
Source: Dân trí Công nghệ, published August 7, 2026. All figures cited are from OpenAI’s internal evaluations as reported by Dân trí.