The new agents are powered by Astra, which OpenAI describes as its latest model. Access to dots is included in the Pro plan without deducting from a user’s standard usage allowance. Alongside the agents, OpenAI introduced “ChatGPT Space,” a collaborative workspace where users can manage pages, presentations, and “living documents.” Pages are described as interactive documents containing charts, images, checklists, and dashboards, allowing ChatGPT to build structured work products based on user requirements. This development positions ChatGPT as a direct competitor to similar agent-based offerings from other major technology firms, aiming to provide a more integrated consumer experience for AI-assisted work.
Image generation capabilities have also been significantly enhanced with the release of ChatGPT Images 2.5. The update focuses on improving the realism of faces, pets, and photo editing tasks, addressing areas where previous models often struggled with consistency and detail. This release is part of a broader push to make the platform’s creative tools more professional and reliable for everyday users.
Alongside the new features, OpenAI is deploying updated core models to all user tiers. Paid Plus and Pro subscribers are receiving a more reliable version of GPT-5.6 Sol, while Free and Go users are transitioning to GPT-5.6 Luna. The primary goal of these updates is to make ChatGPT more direct, factually accurate, and consistent across both simple queries and complex reasoning tasks. A new reasoning slider has been introduced, giving users control over the model’s depth of analysis, ranging from “Instant” for quick responses to “High” for tasks requiring several minutes of processing, such as research, coding, or complex planning.
OpenAI states that GPT-5.6 Sol has been specifically tuned for direct consumer interaction rather than the longer-running agentic workflows found in Codex or ChatGPT Work. The model is designed to adapt its level of detail to the question, avoiding unnecessary formatting and providing direct answers to simple inquiries. For example, a question about the current U.S. president would yield a single-line response, while a question about the mechanics of U.S. elections would provide a more detailed explanation with background information.
In internal testing, OpenAI reported that responses containing at least one factual error were 68% less common with GPT-5.6 Sol compared to the previous GPT-5.5 Instant model. GPT-5.6 Luna, the model for free and lower-tier paid users, showed a 62% reduction in factual errors in the same evaluations. The company claims the new models are less likely to make mistakes regarding dates, numbers, sources, legal rules, medical information, and financial data.
GPT-5.6 Luna is becoming the default model for Free and Go users this week, with some users already reporting the change. Starting next week, OpenAI plans to remove rate limits for text chats on these plans, although usage will remain subject to abuse protections. Limits will continue to apply to file uploads, image generation, and other non-text tools. This move effectively makes unlimited text-based interaction available to the company’s largest user segment, signaling a shift toward broader accessibility rather than strict usage caps for basic functionality.
