OpenAI shipped a meaningful upgrade to ChatGPT this week, and the headline is simple: most of it lands for free users too. The company introduced GPT-5.6 Sol for Plus and Pro subscribers, made GPT-5.6 Luna the default for Free and Go tiers, and gave everyone a couple of new controls over how the model reasons. No paywall required for the biggest change.
What actually changed
The two model variants sit in different lanes. GPT-5.6 Sol is aimed at people who want sharper answers, and OpenAI says it cuts factual errors by roughly two thirds compared to earlier versions. That is a large claimed jump in reliability, the kind of number that usually comes with caveats about the benchmark it was measured on, so treat it as directional rather than gospel. The point is that OpenAI is pushing hard on accuracy for its paid tier.
GPT-5.6 Luna, by contrast, is the workhorse model now serving Free and Go accounts. The notable shift here is access: OpenAI removed the cap on text chats for those users. Previously the free tier throttled how many conversations you could hold in a period. Now you get unlimited text chats, which is a meaningful change for anyone who uses ChatGPT as a daily assistant rather than for occasional queries.
The new controls
Two features shipped alongside the models. The Intelligence Slider lets you dial how much reasoning effort you want from a response, trading a little latency for a lot more depth when a question needs it. There is also a Think button for steering the model toward a more deliberate pass at a hard problem. These are the kind of controls that matter to people who have bumped into the ceiling of a model that answers instantly but shallowly.
None of this replaces the distinction that already exists between tiers, but it narrows it. The free tier gets the reasoning controls and unlimited chats, while paying users get the higher-accuracy Sol model and, presumably, the better multimodal and rate limits they already had.
Why this matters for developers
For anyone building on top of ChatGPT or the underlying models, the accuracy gains in Sol are the more interesting signal. If GPT-5.6 Sol does reduce hallucinations meaningfully, then applications that were previously too unreliable for production, things like data extraction, summarization of critical documents, or agentic workflows where a wrong answer propagates downstream, become more viable. Fewer hallucinations does not mean zero, so keep your evaluation harnesses and human review in place, but it changes the risk calculation.
The unlimited free chats also matter for teams running experiments. When you do not have to meter every free-tier conversation, it becomes far easier to prototype against the model, test prompts at scale, and build internal demos before committing to a paid plan. That lowers the barrier to evaluation, which is almost always a good thing for adoption.
Practical things to check
If you are a ChatGPT user, the fastest way to see what changed is to open the model picker and confirm whether you are on Luna or Sol, then try the Intelligence Slider on a question you have struggled with in the past. Compare a hard reasoning task at low and high effort and see whether the extra latency buys you a materially better answer. If it does, the control earns its place in your workflow.
For developers, treat the accuracy claims as an invitation to re-run your own evals rather than a reason to trust the marketing. Build a small set of representative prompts from your domain, score the new model against the old one, and let the data decide whether you migrate. The upgrade is worth testing regardless of what the announcement says, because accuracy is heavily task dependent and your distribution is not OpenAI’s.
How the tiers stack up now
If you have been deciding between Free, Plus, and Pro, the lines shifted. Free and Go users now get a solid default model plus unlimited text chats, which makes ChatGPT genuinely usable as a persistent assistant without a subscription. That covers a large chunk of what casual users actually need: drafting, research, brainstorming, and iteration over a long conversation without hitting a cap.
Plus adds Sol’s higher accuracy, better rate limits, and the reasoning controls applied to a more capable model. If your work involves factual precision, long documents, or complex multi-step reasoning, the difference between Luna and Sol tends to show up exactly there. Pro sits above that for heavy users who exhaust Plus limits, which is a smaller group in practice.
The practical takeaway: if you only dabble, the free tier is now a lot closer to the paid experience than it used to be. If you depend on ChatGPT for work, evaluate Sol on your own tasks before assuming the price bump is worth it, because the accuracy improvement is real but uneven across use cases.
The broader shape
The move fits a pattern OpenAI has leaned into all year: pushing the frontier for paying customers while expanding the free tier just enough to keep the flywheel turning. Unlimited text chats on the free tier is a real product decision, not a token gesture, because it directly drives the usage data and word of mouth that feed the ecosystem. Whether it survives long term depends on cost, but for now it is a genuine win for users who have been living with the old caps.
The upgrade is worth half an hour of your time, either as a consumer testing the new controls or as a developer re-running evals against Sol. The free-tier changes alone justify a look.