OpenAI GPT-6.1 Terra System CardProvider report

Name: OpenAI GPT-6.1 Terra System Card
Published: 2026-10-26

2026-10-26Twelve GPT-6.1 Terra system-card capability metrics, with same-source GPT-6 Astra comparisons for two new cybersecurity metrics. HealthBench uses length-adjusted scores, retaining unadjusted scores and response lengths. MentalHealthBench uses max and retains standard errors and task counts. ExploitBench uses ACE-normalized capability, separate from arbitrary code-execution success rate; historical vulnerabilities may cause contamination. Sandbox Bench reports successes among 22 targets. Reasoning effort is labeled only when explicit; model metadata and historical reports are unchanged.

Back to data sources

OpenAI GPT-6.1 Terra System CardProvider report

View source

2026-10-26Twelve GPT-6.1 Terra system-card capability metrics, with same-source GPT-6 Astra comparisons for two new cybersecurity metrics. HealthBench uses length-adjusted scores, retaining unadjusted scores and response lengths. MentalHealthBench uses max and retains standard errors and task counts. ExploitBench uses ACE-normalized capability, separate from arbitrary code-execution success rate; historical vulnerabilities may cause contamination. Sandbox Bench reports successes among 22 targets. Reasoning effort is labeled only when explicit; model metadata and historical reports are unchanged.