GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum

OpenAI has introduced GPT-5.1 Instant and GPT-5.1 Thinking, the latest iterations of the GPT-5 family. These models introduce adaptive reasoning capabilities, allowing the model to determine the optimal amount of thinking time based on the complexity of the query, and expanded safety evaluations focusing on mental health and emotional reliance.

Model Capabilities and Routing

GPT-5.1 Instant (gpt-5.1-instant) is designed to be more conversational and features improved instruction following. It employs an adaptive reasoning capability that enables the model to decide when it is necessary to think before responding.

GPT-5.1 Thinking (gpt-5.1-thinking) is optimized for deeper reasoning, adapting its thinking time more precisely to the specific requirements of each question.

To streamline the user experience, GPT-5.1 Auto will automatically route queries to the model best suited for the task, reducing the need for manual model selection by the user.

Safety Mitigations and Evaluation Metrics

While the safety mitigations for the GPT-5.1 series are largely consistent with the original GPT-5 System Card, this addendum provides updated baseline safety metrics.

OpenAI has expanded its pre-deployment safety reviews to include specific evaluations for two new categories:

• Mental Health: Evaluations cover scenarios where a user may exhibit signs of isolated delusions, psychosis, or mania.
• Emotional Reliance: Evaluations focus on outputs related to unhealthy emotional dependence or attachment to ChatGPT.

These updates reflect an expansion of the baseline safety evaluations conducted during the pre-deployment phase to ensure the model's behavior in sensitive conversations.