Publication date: October 8, 2026
OpenAI began rolling out new October versions of GPT-6 Sol and GPT-6 Luna in ChatGPT on October 7. Paid tiers receive Sol, while Free and Go start receiving Luna on October 8. The update replaces GPT-5.6 in ChatGPT and adds model-trained interactive responses and earlier partial answers.
The important distinction is deployment, not a newly named API model. OpenAI’s safety documentation separates these October ChatGPT versions from the GPT-6 models already available in Codex and ChatGPT Work since September.
ChatGPT now decides what an answer should look like
GPT-6 in ChatGPT can assemble text, graphics and interactive elements within one response. OpenAI calls the capability Intelligent UI. It can produce interactive diagrams, comparison layouts and simple tools, such as calculators, without requiring users to open another application.
OpenAI says the model was trained to choose when those interfaces help. A component library and compiler render the interface progressively as it is generated. This is more than a new button in ChatGPT: the October models were trained to decide how to organize and present an answer.
The same rollout also lets GPT-6 start responding before it finishes reasoning or using tools. In OpenAI’s internal tests, GPT-6 Instant begins answering questions requiring web search 44% sooner on average than GPT-5.6 Instant. That is a time-to-first-answer claim, not evidence that every task completes 44% faster.
Free users get GPT-6 Luna, Enterprise waits for its admins
Plus, Pro, Business and Enterprise users start receiving GPT-6 Sol in the Chat tab on October 7. Free and Go begin receiving GPT-6 Luna on October 8. Enterprise availability depends on workspace administrators, so a company may not see the change everywhere at once.
OpenAI explicitly says this release does not change the models powering Codex or ChatGPT Work. Those products continue using previously released September GPT-6 versions. Teams evaluating a ChatGPT conversation should not assume they are testing the same snapshot as their coding agent or a Work workflow.
The standard GPT-6 API model pages still describe the developer products, not an independent public endpoint for the October ChatGPT snapshots. No separate token tariff is attached to this consumer rollout. Access is governed by the relevant ChatGPT plan and rollout status.
Faster starts do not settle the safety or accuracy question
OpenAI reports better web-search decisions and stronger resistance to multi-turn jailbreak attempts than GPT-5.6. Its October safety card also documents statistically significant regressions in certain standard safety tests. GPT-6 Sol regressed on self-harm, while Luna regressed on self-harm, gore and sexual content in the specified evaluations.
OpenAI says its manual review found those failures generally low severity and describes additional safeguards. The company assesses both October models as High capability in cybersecurity and biological or chemical domains under its Preparedness Framework. Those disclosures matter for enterprise governance even if most everyday users never encounter such cases.
Interactive answers add another operational question. A chart, calculator or decision tool can feel more authoritative than prose, but its underlying assumptions still need verification. OpenAI has not published a like-for-like independent evaluation of the new interface experience against the September GPT-6 models.
Test the new answers on real work before they reach your reports
For most organisations this is a change in how staff use ChatGPT, not a new model to build on. Developers have nothing new to call. The people affected are employees who use ChatGPT every day, and the managers who read what they produce.
An answer that arrives as a tidy chart or a working calculator looks more finished than a paragraph, so people check it less. Before Intelligent UI spreads through your team, try it on three or four real tasks: a supplier comparison, a budget calculation, a summary of figures from a document. Check that the numbers in each chart match the source and that each calculator uses the right formula. Enterprise admins can hold the switch until that is done; Business plans get it straight away. If you train staff on AI use, add that checking habit to the programme now, before the first wrong chart ends up in a board pack.
For more information, visit the official announcement of GPT-6 in ChatGPT on the OpenAI blog.
