GPT-6.1 Terra Cuts Single Agent Task Cost to $0.29, OpenAI Starts Competing on Pricing Economics
OpenAI released GPT-6.1 Terra, a model optimized for computer-use and vision capabilities, with API pricing set at $1 per million input tokens and $5 per million output tokens. At a single-task cost of only $0.29, it matches the medium-effort performance score of GPT-6 Astra on the Agents’ Last Exam benchmark, while the equivalent cost for Astra is $4.82.
OpenAI has officially launched GPT-6.1 Terra. Its positioning is very clear: a fast, high-performance model built for computer-use and vision workloads, with API pricing of $1 per million input tokens and $5 per million output tokens. According to MarkTechPost's reporting, Terra scores on par with GPT-6 Astra under the medium-effort setting of the Agents’ Last Exam benchmark, with a single-task cost of just $0.29, compared to Astra's $4.82. The original article headline also notes that Terra cuts the API price of GPT-6.1 Sol in half. While the report body does not expand on Sol's specific pricing, Terra's aggressive price tag is already public information.
For users who do not run AI agent workloads, these terms may feel obscure. Computer-use refers to the model's ability to view screens, click buttons, fill out forms, and operate software on your behalf. Vision refers to its image interpretation capability. Agents’ Last Exam is a dedicated benchmark test built for AI agents, and the "medium-effort" setting can be understood as the tier that does not max out inference resource budget. A token is the standard billing unit for LLM APIs, with input and output tokens calculated separately.
Among this set of figures, the per-task cost is far more worth attention. Benchmark scores determine whether a model is capable of completing a task, while single-task cost determines whether you feel comfortable scaling it to handle more tasks. Anyone who has received an AI agent bill knows well: one single task can involve dozens of rounds of API calls, and even a tiny gap in model efficiency can lead to a full order of magnitude difference in total billing. Terra's single-task cost is $0.29, while Astra's is $4.82, marking a roughly 16x gap. This disparity carries far more meaningful insights than the rollout of a regular new model.
This time OpenAI is not only talking about model capabilities, but also emphasizing the cost accounting of agent workflows. The more skilled a model is at operating computers, the denser its API calls become, and the more inference costs start to feel as routine and ubiquitous as utility bills. Terra is explicitly positioned to let developers get their agent tasks up and running first, before they need to consider whether to upgrade to a more expensive higher-tier model.
Source: MarkTechPost, author Michal Sutter, original article link: https://www.marktechpost.com/2026/10/26/openai-gpt-6-1-terra/
发布时间: 2026-10-27 10:45