Addendum to GPT-6 Astra System Card: GPT-6.1 Terra
OpenAI's system card addendum for GPT-6.1 Terra, rating it Critical in cybersecurity and High in biological and chemical capability and reporting its safety evaluations.
Addendum to GPT-6 Astra System Card: GPT-6.1 Terra
OpenAI Deployment Safety Hub, October 26, 2026. Summary in our own words; original: https://deploymentsafety.openai.com/gpt-6-1-terra
Preparedness classification
OpenAI treats GPT-6.1 Terra as Critical in cybersecurity and High in biological and chemical capability. It runs the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, with probes updated for the new model. Training data and methods are the same as for GPT-6 Astra.
Safety evaluations
Challenging production prompts: on par with GPT-5.6 Terra; small drops on extremism, hate and sexual/minors were not statistically significant.
Under-18 evaluations: better than GPT-5.6 Terra on age-restricted goods, sexual content, eating disorders and gore.
Agentic safe completions: on par with recent GPT-6 and GPT-6.1 releases.
Image inputs: on par for self-harm and erotic content, with statistically significant regressions on hate and extremism; OpenAI says manual review found overall performance in line with prior launches.
Jailbreaks: multi-turn jailbreak robustness on par with GPT-5.6 Terra.
Prompt injection: similar to GPT-6 Astra and GPT-6.1 Sol.
Health and mental health
HealthBench Professional (length-adjusted): 62.7, above GPT-5.6 Terra's 57.7.
Dynamic mental-health simulations: lower than GPT-5.6 Terra on self-harm (0.880 vs 0.947), which OpenAI says manual review found in line with prior launches.
MentalHealthBench overall: 55.5, versus 46.2 for GPT-5.6 Terra.