GPT-6.1 Terra Launches as OpenAI's Fast Work Model

OpenAI has released GPT-6.1 Terra, a fast and efficient model built for the everyday work that fills a business day. According to the announcement reported by The Prompt Insider, the model arrived on October 26, 2026, and is rolling out to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can also call it through the API under the name gpt-6.1-terra.

Terra joins a growing lineup that now includes GPT-6.1 Sol and GPT-6 Astra, which arrived earlier in the autumn. OpenAI positions Terra as the lower-cost, higher-speed option in that family, aimed at high-volume tasks rather than the most complex reasoning jobs. For teams already building with AI agents, that positioning matters, because the cost of each task often decides whether an agent workflow is worth running at scale.

What GPT-6.1 Terra is built to do

OpenAI describes GPT-6.1 Terra as a model that combines detailed visual reasoning with precise computer use and clearer, more natural writing. It is presented as a step up from GPT-5.6 Terra, the previous generation in the same line, for the communication and computer-based tasks that dominate office work.

The practical examples given by OpenAI point to routine but demanding jobs. One is reading incoming invoices and entering their details into accounting software. Another is matching product photos to catalog entries, a task that mixes image understanding with structured data. The model is also said to interpret technical drawings, act on application screens, and draft or revise text.

Those capabilities line up closely with how agents actually work. An agent that can read a screen, fill in a form, and check its own output is far more useful than one that only answers questions. Terra is therefore best understood not as a general chatbot upgrade, but as a workhorse model for agents that operate software on behalf of people. Readers who follow the agent beat will recognize the pattern from recent enterprise launches, including Verovia Industrial AI Brings Agents to Factory Floors, where the value comes from connecting systems rather than from a single clever answer.

Price, availability, and limits

In ChatGPT, GPT-6.1 Terra is available in Work and Codex, not in regular Chat conversations, according to OpenAI's help center as reported by The Prompt Insider. Availability depends on plan, workspace settings, and rollout access, so some users may see it appear over days rather than all at once.

For developers, standard API pricing is set at one dollar per million input tokens, five cents per million cached input tokens, and five dollars per million output tokens. OpenAI says that is fifty percent lower than the pricing for GPT-5.6 Terra and GPT-6.1 Sol. Batch and Flex modes are reported to be fifty percent lower again, while a fast mode costs twice as much.

The model page lists a context window of just over one million tokens and a maximum output of 128,000 tokens, with a knowledge cutoff of June 1, 2026. It accepts text and image input and produces text. Audio and video are not supported, and fine-tuning is not supported either. Tool calling requires the Responses API, with supported tools including web search, file search, image generation, code interpretation, hosted shell, computer use, and the Model Context Protocol.

Those limits are worth noting for agent builders. A large context window helps an agent hold a long task in view, but the absence of fine-tuning means teams must steer the model through prompts, tools, and workflow design instead. The reasoning effort settings, which range from none to maximum, give developers a dial to trade speed against care on each task.

What the benchmarks say, and what they do not

All of the benchmark figures published so far come from OpenAI itself, and no independent evaluations had been found at the time of reporting. That caveat matters, but the direction of the claims is still informative for anyone weighing a switch.

On computer use, OpenAI says Terra at medium reasoning effort scores fifteen percentage points higher than GPT-5.6 Terra on Agents Last Exam, a benchmark of professional tasks, at about one-third of the cost. The company also says Terra performs on par with GPT-6 Astra at one-tenth of the price per token, and completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.

On visual reasoning, Terra is reported to outperform its predecessor at every reasoning setting on MMMU-Pro, and to nearly match the larger models at much lower cost. On desktop tasks, it is said to surpass the best score of GPT-5.6 Terra on OSWorld 2.1 at a fraction of the cost per task. On coding, Terra at maximum effort solves just over half of Terminal-Bench tasks, compared with about one-fifth for its predecessor, although OpenAI still points teams with the hardest coding jobs toward Sol and Astra.

One figure that will interest agent operators concerns factual errors. On an internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from over half with the previous model to under one-fifth with Terra. At maximum effort, it fell further to a low single-digit rate. If those results hold up in production, they would make Terra a safer choice for agents that act without constant supervision, though teams should still test on their own tasks before trusting it with sensitive work.

Safety notes and the bigger picture

OpenAI also published a system card addendum on the same day. Under its Preparedness Framework, the company says it is treating GPT-6.1 Terra as Critical in cybersecurity and High for biological and chemical capability. Terra uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, with probes updated for the new model.

OpenAI says Terra is more likely than its predecessor to stay within scope and to disclose limitations and blockers honestly, and that it showed strong resistance to malicious instructions in prompt-injection tests. The addendum also notes small regressions in a few categories that it describes as not statistically significant, and cautions that evaluations may differ from production ChatGPT because of differences in system prompts, tools, and settings.

For the agent ecosystem, the launch fits a clear trend. The industry is moving from impressive demos toward governed, affordable agents that can be trusted with real workflows, a theme also visible in Bretton AI agents slash bank compliance reviews by 87%, where measurable time savings came from agents working inside strict controls. Terra does not settle the governance question on its own, but a faster and cheaper model that is harder to mislead makes the economics of careful, supervised agents considerably better.

Teams considering Terra should start small: run it on a sample of real invoices, catalog photos, or support documents, compare the output against answers that have already been checked, and measure cost per completed task rather than price per token. That discipline, more than any single benchmark, will show whether Terra earns a place in the agent stack.

Sources: The Prompt Insider, GPT-6.1 Terra price, features, and availability, reporting OpenAI's announcement, API model page, help center guidance, and system card addendum dated October 26, 2026.