GPT-6 Intelligent UI: Interactive Widgets & Sol vs Luna Tiers

TL;DR: OpenAI’s GPT-6 rollout introduces an ‘Intelligent UI’ that replaces static text with interactive widgets, offering Sol to paid users and Luna to free tiers. While this enhances usability, Thai processing costs remain high due to tokenization, highlighting a trade-off between advanced interactivity and linguistic efficiency.

Key facts

• OpenAI globally rolled out GPT-6 starting October 7, replacing GPT-5.6 with ‘Intelligent UI’ that renders dynamic charts and widgets.
• Paid users (Plus, Pro, Business, Enterprise) receive GPT-6 Sol, while free and Go-tier users are assigned the lighter GPT-6 Luna variant.
• GPT-6 Instant, optimized for web searches, was released on October 7 for paid accounts, reportedly starting answers 44% faster than GPT-5.6 Instant.
• Safety updates in October GPT-6 Sol and Luna incorporate advances from the earlier GPT-6 Astra model, strengthening protections against cybersecurity and biological misuse.
• GPT-6 Astra was pretrained on over 100,000 GPUs at OpenAI’s Stargate site in Texas, following delays caused by unsanctioned cyberattacks in July 2026.

OpenAI has completed its global rollout of GPT-6 within ChatGPT, marking a fundamental shift from static text responses to interactive, app-like interfaces [1]. Starting October 7 for paid subscribers and expanding to free users on October 8, this update replaces the previous GPT-5.6 models in the main chat experience [2]. The core innovation is the ‘Intelligent UI,’ which transforms standard paragraphs into dynamic layouts featuring charts, diagrams, tappable buttons, forms, and custom widgets tailored to specific user queries [1]. This approach aims to make AI output more functional for everyday tasks rather than merely informational.

The rollout introduces distinct model variants based on subscription tiers. Paid users, including those on Plus, Pro, Business, and Enterprise plans, receive GPT-6 Sol, while free and Go-tier users are routed to a lighter version known as GPT-6 Luna [4]. For users requiring speed over depth, OpenAI also released GPT-6 Instant on October 7 for paid accounts. This variant is optimized for web searches, reportedly starting answers 44% sooner than the previous GPT-5.6 Instant model [2]. The Intelligent UI operates progressively; text appears instantly to provide immediate context, followed by visual and interactive elements as they are generated, ensuring a smooth user experience even when complex widgets are loading.

Safety remains a central component of this release, with OpenAI’s Deployment Safety Hub noting that the October versions of GPT-6 Sol and Luna incorporate safety advances from the earlier GPT-6 Astra model [4]. These updates include strengthened protections against high-risk misuse in cybersecurity, biology, and violence, as well as improved resistance to multi-turn jailbreaks and reductions in dishonesty [3][4]. OpenAI classifies these October versions of GPT-6 Sol and Luna as ‘High capability’ in Cybersecurity and Biological/Chemical domains under its Preparedness Framework [4]. Despite these improvements, the company acknowledges regressions in certain evaluation areas, characterizing the resulting disallowed responses as generally low severity.

The broader context of GPT-6’s development involves significant infrastructure efforts. OpenAI’s VP of Research Aidan Clark stated that GPT-6 Astra was pretrained on more than 100,000 GPUs at their Stargate site in Texas [3]. The model was initially unveiled as a limited preview on September 3, 2026, following delays caused by unsanctioned cyberattacks involving OpenAI agents in July 2026 [3]. While competitors like Anthropic and Google shipped parallel updates during this period, including a cheaper Claude Haiku 5.5 model, OpenAI’s focus remains heavily placed on making AI output more interactive for everyday tasks [1].

For Thai users, the cost structure of GPT-6 varies significantly by tier, with additional complexity due to tokenization differences in the Thai language. According to KoishiAI’s measurements as of September 28, 2026, GPT-6 Astra costs ฿122 to send 1 million Thai characters (approximately 120 news articles) and ฿612 to receive them [5]. In comparison, GPT-6 Sol is more expensive for input at ฿24.49 per million characters sent, while receiving remains at ฿122 per million characters [5]. The free-tier GPT-6 Luna is substantially cheaper, costing only ฿1.22 to send and ฿6.12 to receive 1 million Thai characters [5]. These figures reflect an exchange rate of 33.345 THB/USD and a tokenization rate of 0.367 tokens per Thai character.

The cost disparity between languages is notable; the same text costs 1.65× more in Thai than in English because GPT-6’s tokenizer splits Thai into more tokens [5]. In terms of input pricing, GPT-6 Luna ranks 6th out of 48 models by Thai input cost on KoishiAI’s tracker, while GPT-6 Sol ranks 37th and GPT-6 Astra ranks 46th [5]. The cheapest model per Thai character remains Qwen3.7 Flash at ฿0.2812 per million characters, highlighting that while GPT-6 offers advanced interactive features, it is not the most cost-effective option for high-volume Thai text processing [5].

Despite the technical advancements, several questions remain unanswered by the available reports. It is unclear how GPT-6 Luna specifically differs in capability or latency from GPT-6 Sol beyond just the naming convention. Additionally, OpenAI has not specified what metrics define ‘low severity’ for the acknowledged safety regressions, nor have they detailed how the Intelligent UI impacts data privacy or processing costs differently for free versus paid users [1]. Competitors like Anthropic’s Claude Haiku 5.5 also lack clear technical comparisons regarding interactive output capabilities in this context.

The shift from text-based AI to ‘app-like’ interactive interfaces marks a significant change in how users consume information, raising questions about accessibility and cognitive load alongside the technical rollout of GPT-6. For Thai readers who may not have English proficiency, the ability to interact with charts and forms directly within ChatGPT could lower barriers to understanding complex data. However, the higher tokenization cost for Thai text means that heavy users may still face significant expenses compared to English-speaking counterparts.

As OpenAI continues to refine GPT-6 across its variants, the focus remains on balancing interactivity with safety and cost-efficiency. The rollout of Intelligent UI represents a move towards more immersive AI experiences, but the long-term impact on user behavior and the economic viability for non-English languages like Thai will depend on how these factors evolve in future updates.

Sources

1. GPT-6’s Intelligent UI in ChatGPT: What It Does and How to Use It (www.mindstudio.ai) — 2026-10-10
2. GPT-6 and Intelligent UI in ChatGPT (community.openai.com) — 2026-10-07
3. GPT-6 Sol and GPT-6 Luna: October 2026 update - OpenAI Deployment Safety Hub (deploymentsafety.openai.com) — 2026-10-07
4. GPT-6 - Wikipedia (en.wikipedia.org) — 2026-09-03
5. AI price per Thai text (KoishiAI) (koishiai.com) — 2026-09-28