OpenAI tightens Sol answers and gives Free users unlimited Luna text chats

On August 6, 2026 OpenAI changed ChatGPT on two layers at once. For Plus and Pro it tunes GPT-5.6 Sol toward shorter, more fact-reliable answers and adds a slider for thinking depth. For Free and Go it moves the default to GPT-5.6 Luna, promises unlimited text chats, and adds a Think button for harder questions.

Sol is supposed to answer more tightly and miss fewer facts

OpenAI says Sol now leads with the main recommendation, scales detail to the question, and avoids unnecessary formatting. The same model is meant to power Instant and deeper reasoning so raising effort feels like more time, not a jump into a different voice.

In an internal evaluation on financial, medical, and legal prompts that need factual detail, responses with at least one factual error were about 62% less common with GPT-5.6 Luna and 68% less common with GPT-5.6 Sol than with GPT-5.5 Instant, according to OpenAI. That is the company's own measurement, not an independent benchmark. The sample bike ride to Ocean Beach is there to show Sol saying "you stay dry" first, then handling wind, instead of dumping a long bullet list.

Free gets Luna while Plus gets depth control

For everyday users the practical change is not the "more intelligence" slogan. Free and Go are to get Luna as default this week. Starting next week: unlimited text chats and a Think button, subject to abuse guardrails. Limits stay on file uploads, images, and other tools.

Plus and Pro get the depth slider on web, mobile, and desktop, and the updated Sol "starting today." OpenAI states clearly that this chat-optimized Sol is only for the Chat experience. The Sol that powers Work and Codex does not change in this release. Anyone building Codex workflows should not expect the same diff from this post.

An internal 68% error drop is not a public audit

The error-rate numbers are strong and still live inside OpenAI. We do not see the dataset, the sampling, or whether users will observe the same drop on their own citation-heavy prompts. The depth slider is a useful UI signal. It does not prove that higher effort actually checks sources instead of spending longer on more confident prose.

Unlimited Free text is real competitive pressure. At the same time OpenAI keeps tool limits and splits chat Sol from Work and Codex Sol. That is product segmentation, not "one model for every job."

Whether Sol stays short and user corrections fall will decide if this stuck

Watch three signals. First, whether Plus and Pro users after a week report fewer "make it shorter" requests and fewer everyday fact corrections. Second, whether Free unlimited text triggers abuse waves and how hard the Think guardrails land. Third, whether chat Sol and Work/Codex Sol diverge enough that teams must name which Sol they actually tested.

Lilith's verdict

OpenAI is not selling a new brain here. It is selling a tighter Sol for paid chat and Luna without a text ceiling for Free, while Codex Sol stays in a different lane.