OpenAI's GPT-6.1 Terra cuts task cost to a third of GPT-5.6 Terra

OpenAI's pitch for GPT-6.1 Terra centers on the economics of getting work done rather than raw capability alone. On Agents' Last Exam, which evaluates professional work tasks, the model delivered a 15 percentage point higher score than GPT-5.6 Terra at roughly one-third the cost under a medium reasoning setting. Task completion speed was more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. On Terminal-Bench 4.0, which measures autonomous command-line work at maximum reasoning, the model solved 50.3% of tasks versus 21.5% for GPT-5.6 Terra — more than double. On the science-focused Terminal-Bench-Science 0.1, it came within 5 points of Claude Opus 5.5 at an estimated cost less than one-fifth of that model. The release also improved factual accuracy on difficult prompts and cleared the Critical threshold in OpenAI's Preparedness Framework for cybersecurity.