OpenAI announced on August 6 that ChatGPT's free tier is about to feel less like a meter that runs out mid-conversation. GPT-5.6 Luna becomes the default model for Free and Go users this week, replacing GPT-5.5 Instant. Starting next week, those same tiers get unlimited text chats, no daily message cap on plain typed conversation, plus a Think button that spends more compute on harder questions. File uploads, image generation, voice, and other tools keep their limits. TechCrunch and The Verge both carried the same core details from OpenAI's product post, including the company's claim that ChatGPT recently crossed about one billion weekly users.
What free and Go users actually get
Luna is the fast, cheap tier of the GPT-5.6 family OpenAI launched in July. Making it the default for Free and Go is the part that lands first. OpenAI says the switch rolls out this week without a settings hunt, the product just answers with Luna unless you dig for something else. Unlimited text chats arrive the week after, subject to what the company calls abuse guardrails. If you have been hitting a wall after a busy afternoon of drafting emails or debugging a script, that wall moves for text-only threads. It does not move for the expensive extras. OpenAI was explicit: limits still apply for file uploads, images, and other tools. Voice stays capped too, according to TechCrunch's write-up.
Free and Go also get a Think button for questions that need more deliberation. It is not the same control Plus and Pro get on Sol. Think is a discrete mode on Luna: tap it when the answer needs more working time. OpenAI has not published a public rate card for how often Free users can smash that button before the guardrails kick in. I would not assume it is infinite just because text chat is.
The Verge noted the familiar pattern: headline generosity on the thing that is cheapest to serve at scale, while multimodal and tool use stay metered. That reads as capacity management dressed up as product generosity, and it is still a real upgrade for people who only ever type.
Go subscribers sit in the same bucket as Free for Luna-default and unlimited text. That is a quiet positioning choice: Go is no longer 'pay to escape the text meter,' it is pay for whatever else OpenAI still meters.
Plus and Pro get a different Sol, and a slider
Paid tiers are not sitting still. Plus and Pro get an updated GPT-5.6 Sol tuned for everyday ChatGPT work, quick questions, web research, advice, planning, writing, decisions, with answers OpenAI describes as more compact and less padded. The company says this ChatGPT-optimized Sol is separate from the Sol that powers Work and Codex. Those surfaces are unchanged in this release. If your team lives in Codex, do not expect Thursday's ChatGPT tweak to alter that stack.
Plus and Pro also get a slider for how much thought Sol puts into an answer. Coverage describes options that climb through higher reasoning effort. Secondary reports have used slightly different label sets for the steps on that slider, so confirm the exact wording in the product UI rather than trusting a screenshot from day one. The updated Sol and the slider were described as available the day of the announcement for paid users, while Free and Go changes stagger across this week and next.
OpenAI cited an internal evaluation against GPT-5.5 Instant: factual errors were about 62 percent less common for GPT-5.6 Luna and about 68 percent less common for the updated GPT-5.6 Sol. That is OpenAI grading OpenAI. Independent third-party numbers were not in the announcement. Still, a company willing to put a percentage on factual error rate is inviting people to notice when the model still invents a citation, so the claim will get stress-tested in public quickly.
One subtle product split deserves a bookmark: ChatGPT Work and Codex keep their existing Sol stack under this announcement. Support threads will fill with people who saw a ChatGPT consumer blog post and assumed their company's Work workspace got smarter overnight.
Internal eval percentages travel well in headlines and poorly in procurement decks. If you are a buyer writing a model card, ask which task suite produced the 62 and 68 percent figures and how factual errors were labeled. Until those answers are public, treat the numbers as directional company claims.
Why remove the free-tier text meter now
ChatGPT's free tier has always been a growth engine with a soft ceiling. Competitors, Google's Gemini, Anthropic's Claude free tier, a pile of open-weight chat UIs, have trained users to expect long sessions without a hard stop. If OpenAI's coverage number of roughly a billion weekly users is in the right zip code, even a small retention lift on free text chats is enormous. Unlimited text is also a bet that Luna's serving cost has fallen enough that the old daily cap was leaving money on the table in the form of users who bounced to another tab.
There is a quieter product message too. Luna as default plus a Think escape hatch looks like OpenAI trying to stop free users from constantly wondering whether they are on the "dumb" model. Instant versus the bigger names was a branding problem as much as a capability problem. Luna is still not Sol, but it is current-generation branding, and the Think button gives a story for hard questions without opening the full paid catalog.
I keep coming back to the limits that remain. Anyone building a free-tier habit around PDF uploads, image edits, or voice notes will still feel the meter. The people who win immediately are students, writers, and developers who live in plain text. That is most of ChatGPT's conversational volume, if not most of its compute bill.
There is also a competitive timing read. Google and Anthropic have spent the year making free tiers feel less punishing. OpenAI's counter is stamina. Long, unbroken text threads are how people draft, debug, and study, and owning that session length keeps the habit loop inside ChatGPT.
What to watch over the next two weeks
Rollouts this size are uneven. Some accounts will see Luna today. Some will still be on Instant over the weekend. Unlimited text next week will probably arrive the same way, region by region, client by client. If your Free account still rate-limits text after the stated window, wait a day before assuming the announcement was vapor. Mobile apps and the web UI have shipped ChatGPT changes on different clocks before.
For Plus and Pro subscribers, the useful experiment is boring: run the same research prompt on the new Sol at a low thinking setting and again with the slider pushed up, and see whether the extra latency buys fewer invented numbers. OpenAI's pitch is reliability on dates, quantities, sources, and rules. Those are exactly the failure modes that make people screenshot ChatGPT in anger.
None of this changes API pricing. Developers building on the API already saw a separate Luna and Terra price cut in late July. This August 6 package is a ChatGPT consumer and seat-holder story. The strategic rhyme is the same, though: OpenAI is spending efficiency gains on access and defaults, not only on a flashier flagship demo. Whether Google and Anthropic answer with their own free-tier meters is the next shoe. They may not need to if their free products already feel uncapped for text. OpenAI just removed one of the sharpest reasons to open a second app mid-draft.
If you run a classroom, newsroom, or helpdesk that leaned on Free ChatGPT, update internal guidance. Unlimited text will raise volume and misuse cases. OpenAI's abuse guardrails will catch some of it. Your Acceptable Use policy still has a job.
- LLMs
