OpenAI is updating GPT-5.6 Sol in ChatGPT with more focused answers, claimed accuracy improvements, and a control for adjusting reasoning effort. At the same time, the company is moving free and Go subscribers toward the smaller GPT-5.6 Luna as their default model, expanding access to everyday chats while reserving Sol for paid users.

The changes separate ChatGPT access more clearly by subscription tier. OpenAI says free users will receive unlimited text conversations with Luna, followed by a new Think button that allows the smaller model to spend more time on difficult questions. File uploads, image generation, and other tools will continue to have limits on free accounts, according to release information reviewed by The Decoder.

For AI builders and enterprise buyers, the update illustrates a growing product trade-off: providers are trying to make lower-cost models useful for routine work while using paid access and controls to differentiate stronger reasoning systems.

GPT-5.6 Sol gets more control and shorter answers

OpenAI says GPT-5.6 Sol now reduces unnecessary detail and excessive formatting. Simple questions should receive direct answers with enough supporting context, while more complex requests should produce fuller responses without burying the main recommendation.

The update also adds a reasoning slider for paying ChatGPT users. Lower settings are intended for ordinary questions, while higher settings target planning, research, and coding. The control is available across web, mobile, and desktop applications. The Decoder reported that a similar slider was already available in ChatGPT Work.

OpenAI’s stated aim is to make fast answers and deeper reasoning feel like different operating modes of one model rather than separate models with noticeably different styles. That could reduce the need for users to manually choose between speed and deliberation, although it also introduces another product control that users must understand.

The change applies to ChatGPT, not every OpenAI product. GPT-5.6 Sol remains unchanged in ChatGPT Work and Codex, according to The Decoder’s account of the rollout.

Free users shift to GPT-5.6 Luna

OpenAI plans to make GPT-5.6 Luna the default for Free and Go users later this week. The company says those users will receive unlimited text chats, with a Think button arriving the following week for questions that need more effort.

That is an access expansion in one sense: free users will be able to conduct more everyday text conversations without hitting the same chat restrictions. But the change also removes the strongest reasoning option from the free experience. Luna can be prompted to reason for longer, yet that does not give it the same underlying capabilities as Sol.

The Decoder characterized Luna as capable for its size but warned that smaller models are generally more vulnerable to errors and reliability problems. It cited an earlier incident involving a previous Instant model that generated a substantially incorrect data analysis compared with a larger reasoning counterpart. That example does not establish how GPT-5.6 Luna will perform, but it highlights the practical risk of treating additional reasoning time as a substitute for a stronger model.

For users doing casual drafting or factual lookups, Luna may be sufficient. The distinction becomes more important for financial analysis, legal research, medical information, coding, and other tasks where a confident mistake can create downstream costs.

Accuracy claims remain vendor-reported

OpenAI’s official announcement describes GPT-5.6 Sol as more accurate and consistent, while The Decoder reported figures from an internal evaluation covering finance, medicine, and law prompts. According to OpenAI, responses containing at least one factual error fell by about 68% for GPT-5.6 Sol and 62% for GPT-5.6 Luna compared with GPT-5.5 Instant.

Those results are not independently verified in the available evidence. The comparison also leaves important questions unanswered, including the evaluation’s sample size, task design, error definitions, and whether the prompts reflect real customer workflows. The figures should therefore be treated as OpenAI’s benchmark claims rather than established measures of reliability.

The same caution applies to the broader promise that the slider will improve user outcomes. A reasoning control may help experienced users allocate more compute to hard tasks, but it could also create a new failure mode if people select a low setting for work that requires careful analysis. The Decoder also reported that users have largely ignored ChatGPT’s model switcher, suggesting that adding controls does not automatically produce better model selection.

What the change means for builders and enterprises

The rollout has immediate implications for teams building around ChatGPT. Free-tier users will increasingly encounter a smaller default model, so workflows designed around the behavior of a stronger model may need new validation. Builders should test not only answer quality but also when Luna’s Think mode is triggered, how often it fails silently, and whether users can recognize when an answer needs review.

For startups and product teams, the split may make cost and quality tuning more explicit. Routine classification, drafting, and customer-support tasks may fit a smaller model, while high-value reasoning, coding, or analysis may justify access to Sol or another stronger system. That can reduce inference costs, but only if routing rules and quality checks are reliable.

Enterprise buyers should also distinguish between ChatGPT’s consumer rollout and the behavior of OpenAI’s other products. Since the Sol update does not extend to ChatGPT Work and Codex, organizations cannot assume that the same model behavior or reasoning controls are available across their deployments.

The key operational question is not whether Luna can reason longer. It is whether the model remains accurate enough for the specific workflow after users accept the free default. Teams should measure error severity, escalation rates, latency, and tool limitations rather than relying on general claims about model intelligence.

What to watch next

The first signal will be how quickly the Luna default and Think button reach Free and Go accounts, and whether OpenAI documents separate limits for the new mode. Independent testing of Luna against Sol on coding, research, and data-analysis tasks will be more informative than the company’s internal comparison alone.

Users and developers should also watch whether OpenAI expands the reasoning slider beyond consumer ChatGPT, changes its paid-plan boundaries, or keeps Sol’s behavior different across ChatGPT, ChatGPT Work, and Codex. Reports of incorrect but confident answers will matter particularly if free users adopt Luna for consequential work.

Finally, adoption of the new controls will reveal whether users want adjustable reasoning or prefer automatic model selection. If the slider is rarely used, OpenAI may need to hide more of the complexity behind task-aware routing rather than exposing additional settings.

Creati.ai perspective

OpenAI’s update is less a single model upgrade than a redesign of the ChatGPT ladder. Sol is becoming a more configurable paid capability, while Luna is being positioned as the broad-access workhorse. That structure can improve economics and make routine use easier, but it increases the importance of communicating model limitations clearly.

For AI teams, the practical lesson is to test the default model users actually receive, not the strongest model shown in a product announcement. GPT-5.6 Luna’s real value will depend on independent reliability results and workflow-level performance, while Sol’s reasoning slider will matter only if it produces measurable gains without adding confusing choices.