Overview
- OpenAI announced GPT-6.1 Terra as a cheaper model tuned for complex desktop automation, visual reasoning, and programming tasks and made it available to paid ChatGPT tiers and the OpenAI API under the id gpt-6.1-terra.
- OpenAI published per-token pricing for the API including $1 per 1 million input tokens and $5 per 1 million output tokens and said Terra’s cost per task is much lower than higher-tier models and some competitors.
- The company presented internal benchmark results showing Terra outperformed GPT-5.6 Terra across visual-reasoning and desktop-automation tests and matched or approached GPT-6 Astra on many measures.
- OpenAI reported large drops in factual-error rates and big gains in programming success on Terminal-Bench 4.0, while noting the highest-difficulty programming problems may still be better handled by GPT-6.1 Sol or GPT-6 Astra.
- OpenAI said Terra inherits GPT-6 Astra’s safety protections, improves resistance to prompt-injection, and meets its stated cybersecurity readiness threshold, a design meant to lower risk for firms adopting automated desktop workflows.