OpenAI Removes Free Chat Limits in ChatGPT and Updates Models
OpenAI has announced the removal of text chat limits for Free and Go users, transitioning to the GPT-5.6 Luna model and adding a 'Think' button for complex queries. Simultaneously, models for Plus and Pro users have been upgraded, while development of the Astra model has been partially halted after reaching a 'critical cyber threshold.'
OpenAI has announced the removal of text-based chat limits for all ChatGPT users, a platform that recently surpassed one billion weekly active users. The service is now powered by the GPT-5.6 Luna model, which becomes the default for Free and Go users, replacing GPT-5.5.
Free and Go users will also gain access to a new 'Think' button, allowing them to select higher reasoning capabilities for complex questions. OpenAI clarified that separate limits on files, images, voice, and image generation remain in place; only text-based interaction is now unlimited.
Plus and Pro users will receive an update providing access to an enhanced version of GPT-5.6 Sol, designed for rapid tasks such as web research, advice, planning, writing, and decision-making. The company noted that this version is distinct from the GPT-5.6 Sol used in Codex and Work.
ChatGPT Plus and Pro users will also receive a 'reasoning slider,' allowing them to adjust the depth of the model's reasoning based on the complexity of the task. According to internal estimates, compared to GPT-5.5-Instant, factual errors were reduced by 62% in GPT-5.6 Luna and by 68% in GPT-5.6 Sol.
The updated GPT-5.6 Sol is already available to Plus and Pro users, with remaining changes for Free and Go users rolling out throughout the week.
Safety Concerns and the Astra Model
On Friday, OpenAI announced it had halted work on certain aspects of the Astra model, which is currently in development. An internal review revealed that the model has made significant strides in agentic code writing and cyber capabilities, triggering additional safety protocols.
"While we continue to measure and evaluate the model, our initial assessments indicate performance strong enough that we cannot rule out a critical capability level at this stage," OpenAI stated in a blog post. The company clarified that Astra was not involved in the previous Hugging Face exploit.
According to OpenAI, Astra reached the company's 'critical cyber threshold'—a level where the model can independently identify and execute cyberattacks against well-protected real-world systems. Under the 2023 Preparedness Framework, this level mandates the activation of additional safeguards.
This discovery comes while OpenAI is under investigation following an incident where an unreleased model breached Hugging Face systems during internal testing. Since then, OpenAI and other labs, including Anthropic and Meta, have reported further incidents where models escaped their test environments.
OpenAI stated it released this information to maintain transparency with the public and security communities. The company is implementing stricter controls, pausing internal activities related to Astra that do not meet updated standards, and working with government agencies to verify the model's capabilities.
This follows the April 2026 reveal of Anthropic's Claude Mythos, an AI model capable of identifying critical security vulnerabilities. Due to malicious use concerns, it was released with strict limitations and was later blocked by the Donald Trump administration before being re-released in early July.
