OpenAI’s Terra update arrives with a clear message: more work, lower cost
OpenAI has officially unveiled GPT-6.1 Terra, a new model in the GPT-6 series designed to combine detailed visual reasoning, precise computer use, and clearer natural-language writing for workplace tasks . The company is presenting Terra not as a new top-end flagship, but as a cost-performance model meant to make agentic work more practical at scale .
That positioning matters. In OpenAI’s current model lineup, Terra sits in the middle of the range: more capable than lighter high-volume options, but cheaper than the most powerful models used for complex reasoning and coding. OpenAI says GPT-6.1 Terra is a significant step up from GPT-5.6 Terra, particularly for tasks that involve looking at images or screens and then taking action across software .
The company’s launch materials point to use cases such as drafting proposals, interpreting technical drawings, filling out forms, reading invoices, entering details into accounting systems, and editing videos . In other words, Terra is built for the layer of work where AI is expected not only to answer questions but also to understand interfaces, documents, and visual context well enough to move a task forward.
What OpenAI says GPT-6.1 Terra is built to do
The most important claim in the launch is that Terra connects visual understanding with action. OpenAI describes the model as capable of reading what is on screen, reasoning about that information, and helping with workflows across browsers, desktops, and apps . ChatGPT’s product documentation says users can apply Terra to tasks such as matching product photos to catalog entries, reviewing campaign assets, checking cropping and text placement, and organizing the resulting edits .
That focus makes GPT-6.1 Terra a model for operational work rather than only conversational assistance. For a retailer, the value proposition might be image-to-catalog matching. For an architecture or engineering team, it might be interpreting technical drawings. For a finance department, it could mean extracting invoice details and moving them into a system of record. These examples all share the same pattern: visual input, structured reasoning, and execution in software.
OpenAI also emphasizes writing quality. The company says Terra produces clearer and more natural writing than its predecessor, while ChatGPT’s own “what’s new” notes describe the model’s style as direct, result-first, and plain-language in its explanations . That is a notable part of the pitch because many business workflows require both task execution and a readable summary of what was done.
Performance claims: better computer use, factuality, and coding
OpenAI’s benchmark claims are ambitious. On Agents’ Last Exam, a professional-task benchmark that includes activities such as turning architectural drawings into a 3D model and reviewing animated film shots, OpenAI says GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort while running at about one-third the cost .
The company also says Terra performs on par with GPT-6 Astra at one-tenth the price per token on that benchmark, and that at maximum effort it is over three times as token-efficient as Claude Opus 5.5 while costing one-fourth as much per token, producing an estimated ninefold lower cost per task . These are OpenAI’s own measurements, so they should be read as launch-claim benchmarks rather than independent market verdicts.
On visual reasoning, OpenAI says GPT-6.1 Terra outperforms GPT-5.6 Terra across reasoning settings on MMMU-Pro and nearly matches GPT-6 Astra and GPT-6.1 Sol at much lower cost . On OSWorld 2.1, which measures desktop-app task completion, OpenAI says Terra surpasses GPT-5.6 Terra’s best score at a fraction of the cost per task .
The model is also being pitched as a factuality upgrade. OpenAI says that on an internal set of difficult, error-inducing prompts, the factual error rate at low reasoning effort dropped from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra . At maximum effort, OpenAI says the error rate fell to 4.3%, close to GPT-6 Astra and GPT-6.1 Sol in the company’s internal evaluations .
Coding is another area where OpenAI is claiming a large jump. On Terminal-Bench 4.0, the company says GPT-6.1 Terra solves 50.3% of tasks at maximum reasoning effort, more than double GPT-5.6 Terra’s published 21.5% score . OpenAI still says GPT-6.1 Sol and GPT-6 Astra remain the better choices for the most complex coding tasks, which helps define Terra’s role: cheaper and strong enough for many scoped engineering tasks, but not the preferred model for the hardest code work .
Pricing and availability
The cost story is central to this release. OpenAI says GPT-6.1 Terra’s standard API prices are 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol . Standard API pricing is listed at $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens .
That pricing makes Terra a notable part of OpenAI’s broader cost-intelligence strategy. A cheaper model with strong visual and computer-use skills can change the economics of automation because many business workflows require repeated, high-volume actions rather than a single complex answer. If Terra can perform reliably on routine screen and document tasks, its lower token costs could make it easier for developers to build agents that run continuously or process large queues of work.
OpenAI’s API changelog records GPT-6.1 Terra as released on October 26 for the Responses API and Chat Completions API, describing it as a fast, efficient model for visual reasoning, precise computer use, and clear writing . ChatGPT Learn says Terra is available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu plans, though availability may depend on workspace controls and product setup requirements .
The product documentation also adds some limits that are important for users. Terra can be selected from the Advanced model list in the ChatGPT desktop app or ChatGPT Work on the web, and Work supports Terra on mobile . The same page says Terra is not available in Chat and does not support UAE inference residency . For enterprises, administrators may control access, and browser use or computer use may require additional setup .
Safety: powerful enough to require serious controls
OpenAI released a system-card addendum for GPT-6.1 Terra on October 26, giving additional safety and evaluation details for the model . In that document, OpenAI says it is treating GPT-6.1 Terra as Critical in cybersecurity and High in biological and chemical capability under its Preparedness Framework .
That does not mean OpenAI is saying Terra can automatically carry out every dangerous task. The system card says the model uses the same safeguard stack as GPT-6.1 Sol and GPT-6 Astra, with updated probes compatible with Terra . OpenAI also states that its internal report informed the Safety Advisory Group’s recommendation and leadership’s determination that safeguards were sufficient for public launch .
The cybersecurity discussion is especially significant because Terra is a model designed for computer use. OpenAI says Terra reached a 24.5% arbitrary code-execution success rate on an internal ExploitBench port, compared with 21.5% for GPT-6.1 Sol and 31.5% for GPT-6 Astra . On Sandbox Bench, OpenAI says Terra succeeded on 5 of 22 targets, compared with 9 of 22 for GPT-6.1 Sol and 10 of 22 for GPT-6 Astra . These details underline the tension in the release: the same computer-use capability that makes Terra valuable for workplace automation also demands careful access controls and monitoring.
OpenAI also reports improvements in areas such as under-18 safety evaluations, prompt-injection resistance, and model honesty about limitations . For business users, the practical takeaway is that Terra should be evaluated not only for performance and price, but also for how it behaves in tool-connected workflows where a mistaken or manipulated action could have real consequences.
How the launch is being received
Early coverage has focused on the model’s role as a middle-tier upgrade rather than a frontier-model reveal. 9to5Mac described GPT-6.1 Terra as OpenAI’s new medium-sized model balancing intelligence and affordability, noting that Terra had been missing when other GPT-6 models were updated earlier and is now moving from GPT-5.6 to GPT-6.1 .
Developer-facing discussion also highlights the cost angle. OpenAI’s developer community announcement framed Terra as a model that lets usage allowances go further in ChatGPT Work and Codex, while still pointing users to GPT-6.1 Sol or GPT-6 Astra for deep research or complex coding . That framing is important because it tells developers how OpenAI wants the model to be used: as a dependable workhorse, not as the universal best model.
AI Understanding’s report similarly summarizes the launch around availability in ChatGPT Work, Codex, and the OpenAI API, emphasizing Terra’s role in visual reasoning and computer-use workflows . Across the first wave of coverage, the release is being interpreted less as a pure intelligence milestone and more as a pricing, efficiency, and workflow-automation milestone.
Why GPT-6.1 Terra matters
GPT-6.1 Terra matters because the next phase of AI competition is not only about the smartest possible model. It is also about which models can do useful work affordably, repeatedly, and safely. OpenAI’s launch materials are built around exactly that claim: Terra is meant to be fast enough, capable enough, and cheap enough to make everyday automation more practical .
For developers, the model creates another option in the trade-off between cost and capability. For companies, it may reduce the price barrier to deploying visual and computer-use agents across routine operations. For users, it could make ChatGPT Work and Codex feel less constrained when handling repeated tasks, because OpenAI says the same usage allowance covers more work with Terra than with GPT-5.6 Terra, GPT-6.1 Sol, or GPT-6 Astra .
The unanswered question is how well OpenAI’s benchmark claims translate into live workflows. Real-world computer use is messy: websites change, forms behave unpredictably, permissions interrupt sessions, and visual cues can be ambiguous. Terra’s launch gives OpenAI a cheaper and more specialized tool for those problems, but users will still need to test it against their own processes.
For now, the current state of the story is clear. OpenAI has launched GPT-6.1 Terra, made it available across ChatGPT Work, Codex, and the API, priced it aggressively below higher-tier models, and published safety documentation that treats the model’s computer-use power as significant enough to require robust safeguards . Terra is not being sold as the new ceiling of AI intelligence. It is being sold as the model that can bring more of that intelligence into everyday work.