Visible reasoning outcomes on the MMMU-Professional benchmark present GPT-6.1 Terra outperforming its predecessor, GPT-5.6 Terra, whereas practically matching top-tier frontier fashions like GPT-6 Astra and GPT-6.1 Sol throughout each reasoning configuration, at a fraction of the price. At max effort, GPT-6.1 Terra achieved an 85.5% rating at $0.012 per activity, in comparison with GPT-5.6 Terra’s 81.7% at $0.079. Throughout the present household lineup, GPT-6 Luna scored 79.9% ($0.0021 per activity), GPT-6.1 Sol hit 86.3% ($0.013 per activity), and flagship GPT-6 Astra led at 86.7% ($0.065 per activity). OpenAI didn’t embrace competitor mannequin comparisons for this particular benchmark.
On the OSWorld 2.1 benchmark, which evaluates how AI brokers navigate desktop functions to carry out complicated multi-step duties, GPT-6.1 Terra scored 74.2% at max effort, costing $1.19 per activity. That marks a considerable improve over GPT-5.6 Terra’s 57.1% rating at $4.76 per activity. Nonetheless, it nonetheless trails Anthropic’s Claude Sonnet 5.5, which led the class at 83.9% with a value of $5.74 per activity.
OpenAI additionally claims to have invested closely in lowering hallucinations and factual errors produced, and as per the inner benchmark of error-inducing prompts, the error price at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, and at most effort, the error price fell by 18 share factors to 4.3%, practically matching the charges for GPT-6 Astra and GPT-6.1 Sol. On cybersecurity, GPT-6.1 Terra meets the Important threshold beneath OpenAI’s Preparedness Framework and retains the safeguards developed for GPT-6 Astra.
