GPT-6 ChatGPT rollout expands: first check which model runs on free and paid plans

OpenAI broadly launched a new GPT-6-based ChatGPT experience on October 7. Paid plans default to GPT-6 Sol, while Free and Go plans default to GPT-6 Luna. Since the announcement was made 11 days ago, this is not breaking news. However, developers still need to check that the model in the ChatGPT chat interface differs from the models used in Codex and Work.

1. Intelligent UI has been added to the chat interface

The key point of this update is the response format rather than the model name. OpenAI explains that GPT-6 was trained to choose and use text, images, buttons, forms, charts, and interactive elements according to the question. Comparison questions may appear as side-by-side tables, while explanations of principles may appear as diagrams whose values can be changed. It also includes the ability to request small tools, such as a savings calculator or a simple game, within a conversation.

OpenAI stated that it used its own native component library together with a compiler that gradually draws the screen while generating a response. The company said ChatGPT has 1.2 billion weekly users and introduced this feature as a change designed for that scale. The explanation of the structure is based on the company's announcement, and no independently verified material from an external organization was found.

2. Free Luna and paid Sol have different rollout dates

According to the official announcement, the allocation is as follows. Plus, Pro, Business, and Enterprise users have used GPT-6 Sol by default since October 7. Free and Go users have used GPT-6 Luna since the following day, October 8. Enterprise availability varies according to the workspace administrator's settings. This announcement is limited to the Chat tab, and the company wrote that the models used in Work and Codex would not change.

3. Read the speed and safety figures as company claims

OpenAI stated that, for questions requiring web search, GPT-6 Instant begins answering an average of 44% faster than GPT-5.6 Instant. It also said that GPT-6 more often identified the key issues in difficult problems than GPT-5.6. Both results come from internal company evaluations, and they cannot be reproduced externally because the evaluation tasks and scoring methods were not disclosed.

An update to the system card stated that GPT-6 Sol and Luna are treated as having High capability in the biology and chemistry domains. It says that the biology-related refusal evaluation improved substantially over GPT-5.6 in August 2026. In contrast, it only says that the cybersecurity safety scores were broadly similar to those of GPT-5.6 Sol and Luna in August of the same year. This document provides no evidence of improvement in the cybersecurity domain.

4. Three things to check

First, check the plan you are using. Even when using the same ChatGPT, Free accounts default to Luna and paid accounts default to Sol. If the quality of results created by a team differs, it is appropriate to suspect a plan difference before a model difference.

Second, separate chat results from development-tool results. Intelligent UI changes the response format in the chat interface, while the models in Work and Codex did not change as a result of this announcement. Impressions from chat should not be transferred directly to evaluations of code-generation quality.

Third, inspect the inputs and calculation formulas of interactive outputs separately. Charts and calculators may look good, but incorrect inputs produce incorrect results. For decisions involving money or health, it is safest to use them only after comparing them with the original data.

5. Related article and sources

When choosing a model, you should consider pricing conditions along with performance. Previous article: Gemini 4 Argon Google Cloud launch: calculate the unit price from the November 30 end date for the introductory price ↗

Sources: OpenAI, GPT-6 and Intelligent UI for everyone (October 7, 2026) ↗, OpenAI, GPT-6 Sol and GPT-6 Luna: October 2026 update (system card) ↗

This English edition was written and translated with AI assistance, with sources provided. For errors or additional information, please request a correction. Editorial policy