GPT-6.1 Terra: benchmark results

Provider: OpenAI. Released 2026-10-26. Access: API.

Aggregate Rating 1948 ± 14, rank #10 of 1657 rated models, from 180 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA GDP.pdf - Finance/Investing - Criterion Pass Rate92.74Criterion Pass Rate (%)100
Coarena1170.6Elo (Bradley-Terry)100
Coarena - Task Completion96.4Completion Rate (%)100
OpenAI GPT-6 Astra System Card - Cyber Safety - Synthetic Agentic Environments1Safety score (0-1)100
OpenAI GPT-6.1 Terra System Card - Codex Deployment Simulation (October 2026) - Severity 1-2 Deception0.06Flagged trajectories (%; 48,690 matched tasks)100
OpenAI GPT-6.1 Terra System Card - Codex Deployment Simulation (October 2026) - Severity 1-2 Ignoring Instructions11.8Flagged trajectories (%; 48,690 matched tasks)100
OpenAI GPT-6.1 Terra System Card - Codex Deployment Simulation (October 2026) - Severity 1-2 Reward Hacking0.03Flagged trajectories (%; 48,690 matched tasks)100
OpenAI GPT-6.1 Terra System Card - Codex Deployment Simulation (October 2026) - Severity 3+ All Tasks0.05Severity 3+ flagged trajectories (%; 48,690 matched tasks)100
OpenAI GPT-6.1 Terra System Card - Coronavirus--ACE2 Cell-Entry Screen0.47Composite score (higher is better)100
OpenAI GPT-6.1 Terra System Card - ExploitGym Honeypot - Successful Exploit0Successful honeypot exploit rate (%; no production safeguard100
OpenAI GPT-6.1 Terra System Card - ExploitGym Honeypot - Unsuccessful Attempt0.07Unsuccessful honeypot attempt rate (%; no production safegua100
OpenAI GPT-6.1 Terra System Card - Phage--plasmid Co-evolution10.82Negative log-likelihood (lower is better)100

Interactive version: theaggregate.ai/model?slug=gpt-6-1-terra · How It Works · Data refreshed daily, snapshot 2026-10-27.