OpenAI Lifts ChatGPT Chat Limits for Free Users, Introduces Teen Safeguards

Post Views: 7

OpenAI has released updates to its GPT-5.6 Sol model, expanding access for free users and improving protections for younger audiences.

Expanded Access for Free Users

OpenAI has eliminated text conversation rate limits for free users, enabling uninterrupted interactions without waiting for cooldown periods. Free users will transition to GPT-5.6 Luna as their default model this week. Starting next week, unlimited text chats will be available, alongside a dedicated Think button for complex queries.

Improved Protections for Younger Audiences

GPT-5.6 Sol incorporates additional protections to mitigate risks associated with adolescent users. The model is trained to avoid content involving romantic roleplay, age-inappropriate challenges, and scenarios where it might replace human relationships. Age-specific boundaries now address topics such as sexual material, eating disorders, body-image concerns, dangerous activities, and graphic violence.

Internal testing indicates improved accuracy with the updated model. Evaluations on financial, medical, and legal prompts showed a 62% reduction in factual errors with GPT-5.6 Luna compared to the prior GPT-5.5 Instant model, and a 68% decrease with GPT-5.6 Sol.

Enhanced Model Capabilities

For paid subscribers, GPT-5.6 Sol now integrates both rapid responses and extended analytical processes within a single model, replacing the previous division between Instant and Thinking modes. This change aims to produce more focused answers, adjust response depth based on query complexity, and minimize unnecessary formatting. A new interface element allows users to control the level of reasoning effort applied to each response.

Safety and Compliance Measures

OpenAI reinforced safeguards with system-level checks and introduced new evaluation protocols to assess model performance for users under 18. Testing on jailbreak scenarios showed GPT-5.6 Sol performed comparably to earlier versions like GPT-5.5 Thinking. For prompt injection attacks, GPT-5.6 Sol achieved a 1.000 score on connector evaluations, surpassing the 0.998 score of GPT-5.4 Thinking. On separate tests targeting search and function-calling vulnerabilities, Sol scored 0.910.

Additional updates include revised training to reduce harmful outputs and ongoing refinements to ensure compliance with safety standards. The company continues to monitor and improve model behavior for all user demographics.