Free users of OpenAI's generative artificial intelligence (AI) service, ChatGPT, can also use the latest AI model without limitation.
OpenAI announced that it will provide unlimited text conversations with “Luna,” a lightweight version of its latest AI model “GPT-5.6,” to free users and users of its low-cost “Go” plan. Unlimited conversation is available only through text, while file uploads, image generation, and voice conversations remain subject to token limits as before.
GPT-5.6-Luna is the fastest model among the three GPT-5.6 models including “Sol” and “Terra,” although its performance is relatively lower. In a performance benchmark by AI performance analysis firm Artificial Analysis, it was evaluated as equal to Google's latest lightweight model “Gemini 3.6 Flash” and higher than Anthropic's “Claude Haiku 4.5.”
In addition, it added a “Think” option that allows setting longer reasoning time when asking questions, enabling its use for high-difficulty questions.
“The unlimited offering of the GPT-5.6 Luna model is a step toward abundant intelligence,” OpenAI explained. “It is our mission to ensure that artificial general intelligence (AGI) benefits all of humanity.”
This is interpreted as a strategy by OpenAI, which has been strengthening its personal customer-centric market dominance, to maintain a competitive edge in the business-to-consumer (B2C) market at a time when ChatGPT's weekly active users (WAU) exceeded 1 billion. Introducing advertisements to free and low-cost plans is also expected to expand ad revenue.
For subscribers of paid plans costing $20 or more per month, such as Plus and Pro, OpenAI also improved the top-tier version of GPT-5.6, the “Sol” model, to present clear answers and significantly enhance the reliability of factual relationships such as numbers, dates, and sources.
Late last month, it set off a price-cutting competition in AI models by reducing developer API usage fees by up to 80%. As global competition among Anthropic, China's Moonshot AI, Alibaba, and others intensifies, this is aimed at strengthening the competitiveness of its own services. It is a strategy to continuously retain individual and enterprise customers by expanding accessibility to GPT models and enhancing usability.
Meanwhile, OpenAI decided to postpone the launch of “Astra,” a next-generation AI model that has achieved results in solving difficult problems such as mathematics that humans have failed to solve for more than 10 years.
Earlier internal evaluation results assessed that the next-generation model “Astra” showed significant progress in coding and cybersecurity, potentially reaching the highest grade, “Critical,” according to its internal criteria. This means it has reached a level capable of autonomous attack by identifying zero-day vulnerabilities in multiple critical systems with robust security without human intervention.
OpenAI stated that although the evaluation of Astra has not yet been completed, preliminary evaluation results showed performance was so outstanding that a “Dangerous” rating could not be ruled out, and it decided to temporarily suspend internal activities that do not meet enhanced security standards. According to Axios, amidst this judgment, OpenAI also notified the White House of the decision to postpone the launch of Astra.
