San Francisco:OpenAI has reportedly introduced GPT-6.1 Terra, an artificial intelligence model designed to make workplace automation more affordable through lower API pricing, faster task completion and improved performance on professional workflows.
According to the supplied report, the model combines visual understanding, computer-use capabilities and writing improvements to help developers and businesses automate tasks such as processing invoices, drafting proposals, interpreting technical drawings and completing digital forms.
The reported launch centres on a significant price reduction. Standard API rates are said to be 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol, potentially making the model an option for organisations that use AI to handle large volumes of repetitive administrative work.
However, the model name, pricing schedule and benchmark figures should be checked against OpenAI‘s official documentation before publication.
The reported API pricing is one of the main highlights of the announcement. The model is said to cost $1 per million input tokens, with cached input tokens priced at $0.05 per million tokens. Output tokens are reportedly priced at $5 per million tokens.
| API usage | Reported price per million tokens |
|---|---|
| Input tokens | $1 |
| Cached input tokens | $0.05 |
| Output tokens | $5 |
Input tokens generally represent the text, instructions, documents or other information supplied to a model. Output tokens represent the content it generates in response. Cached input pricing can apply when eligible portions of a prompt are reused, potentially reducing costs for applications that repeatedly process similar instructions or reference material.
The difference between input and output pricing is important when estimating the cost of deploying an AI system. An application that processes large documents but produces short summaries may have a different cost profile from one that generates lengthy reports or proposals.
The supplied report also claims that standard API pricing is 50% lower than that of GPT-5.6 Terra and GPT-6.1 Sol. If confirmed, the reduction could make it more economical for companies to run AI-powered workflows at scale.
What tasks can the model automate?
The reported capabilities of GPT-6.1 Terra focus on business applications that involve multiple steps, different types of information and interaction with software.
- Invoice processing: Extracting invoice details and transferring them into accounting systems.
- Business writing: Preparing first drafts of proposals, reports and other professional documents.
- Technical document analysis: Interpreting information contained in drawings and visual documents.
- Digital form completion: Entering information into online forms and business applications.
- Workflow automation: Performing a sequence of actions across supported software tools.
- Video editing assistance: Helping with editing tasks through compatible computer-use workflows.
These applications could reduce the amount of repetitive work required from employees. For example, a finance team could use an AI system to identify invoice numbers, dates, supplier details and payment amounts before transferring the extracted information into accounting software.
Such workflows would still require appropriate checks. Errors in financial records, incorrect form submissions or changes made to business systems can have practical consequences, so organisations would need to establish permissions, review procedures and safeguards.
Computer-use capabilities and workplace automation
The supplied report highlights computer use as one of the model’s key capabilities.
Computer-use systems are designed to interact with software interfaces rather than merely generate text. Depending on the tools and permissions available, an AI agent may be able to navigate an application, enter information, select options and carry out a multi-step workflow.
This approach could help automate tasks that traditionally require employees to move between different applications. A system might, for instance, extract details from a document, enter them into a business platform and prepare a summary for a human reviewer.
The usefulness of such automation depends on more than the model itself. Performance can also be affected by application design, tool integration, network conditions, access permissions and the complexity of the task.
For businesses, the key consideration is whether the complete workflow is reliable and cost-effective, rather than whether an AI model performs well on an isolated benchmark.
Benchmark results and performance claims
According to the supplied report, OpenAI evaluated GPT-6.1 Terra using Agents’ Last Exam, described as a benchmark for assessing an AI system’s ability to complete professional tasks using computers.
The report claims that GPT-6.1 Terra achieved a score 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort while completing tasks at approximately one-third the cost.
It also claims that the model performed comparably to GPT-6 Astra while costing one-tenth as much per token. The report further states that GPT-6.1 Terra completed benchmark tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
These figures would be significant if confirmed, but meaningful comparisons require consistent testing conditions. Factors such as reasoning settings, task selection, tool access, evaluation methods and the definition of cost can influence benchmark outcomes.
The figures should therefore be checked against the original evaluation results before being treated as established performance measurements.
OpenAI’s reported focus on reducing hallucinations
The supplied report also says GPT-6.1 Terra produces more reliable responses and reduces AI hallucinations.
Hallucinations occur when an AI system generates information that is inaccurate, unsupported or fabricated. They remain a concern in professional applications, particularly when models handle financial records, legal documents, technical information or other material where mistakes can be costly.
Improvements in reliability could make an AI system more useful for workplace automation. However, a reduction in hallucinations does not mean that a model is error-free.
Businesses deploying AI in important workflows would still need validation procedures, audit trails and human oversight. For tasks involving payments, sensitive information or irreversible actions, additional safeguards would be particularly important.
What lower API costs could mean for businesses
If the reported pricing and performance figures are accurate, GPT-6.1 Terra could make it easier for organisations to assess AI automation for high-volume operations.
Lower input costs may benefit applications that repeatedly analyse documents or process large amounts of text. Reduced overall task costs could also make automation viable for workflows that were previously too expensive to operate at scale.
However, API pricing alone does not determine the total cost of deployment. Businesses also need to account for output tokens, software integration, monitoring, data security, error correction and human review.
A model that costs less per token may not necessarily be the cheapest option for every task. Organisations would need to compare the cost of completing a workflow successfully, alongside speed, accuracy and operational requirements.
What to verify before deployment
Before adopting the model, developers and businesses should confirm its official API availability, supported inputs, computer-use tools, pricing conditions and performance documentation.
They should also test it against their own workloads rather than relying exclusively on general benchmark results. A model that performs well on one evaluation may behave differently when processing a company’s documents or interacting with its software.
For now, the supplied announcement presents GPT-6.1 Terra as a model focused on affordable, computer-assisted workplace automation. Its reported combination of lower pricing, faster execution and improved professional-task performance could be relevant to businesses seeking to expand their use of AI.
The model’s actual advantages, however, will depend on confirmation of the specifications and how reliably it performs in real-world applications.
