The headline: cheaper intelligence for routine professional work
OpenAI has launched GPT-6.1 Terra, a new model in the GPT-6 line that is positioned less as a maximum-intelligence flagship and more as a practical workhorse for high-volume professional tasks . The company says Terra combines detailed visual reasoning, precise computer use, and clearer natural-language writing, with standard API token prices that are 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol . That pricing claim is the core of the release: OpenAI is trying to make more capable agentic workflows economically viable when users and developers are running many tasks, not just a handful of premium prompts .
The model is available through the OpenAI API as gpt-6.1-terra, and OpenAI’s API changelog lists it for both the Responses API and Chat Completions as of October 26, 2026 . In ChatGPT products, Terra is being added to ChatGPT Work and Codex rather than the ordinary Chat surface, with the company describing it as a model for professional workflows that involve images, screens, apps, browsers, and documents . That distinction matters: Terra is not being sold as the new default chatbot for every conversation, but as a lower-cost engine for doing work.
What changed in pricing
The standard API price is the simplest part of the story. OpenAI lists GPT-6.1 Terra at $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens . Because GPT-6.1 Sol is priced higher, Terra gives developers a cheaper option when Sol’s extra reasoning depth is not necessary . AWS’s launch note makes the same cost positioning, saying Terra’s input and output token prices are at least 50% lower than GPT-6.1 Sol and GPT-5.6 Terra .
This does not mean every real-world bill will fall by exactly half. The cost of an agentic workflow depends on the number of turns, tool calls, screenshots, cached context, retries, and reasoning effort. Still, a 50% token-price reduction changes the economics of workflows that were previously marginal: invoice intake, form completion, catalog matching, codebase triage, document review, and internal operations agents. These are tasks where teams may care less about absolute frontier reasoning and more about stable, repeatable performance at a predictable price.
Why Terra is not simply a “cheap model”
OpenAI’s framing is careful: Terra is cheaper, but it is not presented as a stripped-down assistant. The launch post says Terra is a significant step up from GPT-5.6 Terra, especially for the everyday communication and computer-based tasks that fill a workday . The examples are practical rather than theatrical: drafting proposals, interpreting technical drawings, filling out forms, and editing videos .
The strongest positioning is around visual understanding followed by action. OpenAI says Terra can reason about what appears on a screen and then act across apps and browsers . That combination is important for agents because many business tasks are not pure text generation. They involve reading interfaces, extracting structured details, clicking through systems, and producing a usable result. A lower-cost model that can perform those steps reliably can be more valuable than a more powerful model that is too expensive for routine deployment.
Performance claims and where to be cautious
OpenAI reports that GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra on Agents’ Last Exam at medium reasoning effort, while costing about one-third as much on that benchmark . The company also says Terra performs on par with GPT-6 Astra on that test at one-tenth of Astra’s token price, and that it completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol . AWS repeats the same broad performance framing in its Bedrock availability announcement, including Terra’s stronger result over GPT-5.6 Terra and its faster task completion versus Astra and Sol .
Those numbers are important, but they should be read as vendor-reported launch benchmarks. They help explain how OpenAI wants buyers to think about Terra, but organizations should still run their own evaluations on the actual tasks they plan to automate. A model that excels at reading technical drawings or browser screens may not be the right choice for every deep coding, research, legal, financial, or scientific task. OpenAI itself says that for the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice .
Coding, factuality, and practical reliability
The release is also a coding story, though not a pure coding story. OpenAI says GPT-6.1 Terra solves 50.3% of tasks on Terminal-Bench 4.0 at maximum reasoning effort, more than double GPT-5.6 Terra’s published score of 21.5% . That is a meaningful jump if Terra is being used for terminal work, codebase investigation, scoped bug fixes, and repetitive development tasks.
OpenAI also claims a large improvement in factuality. On its internal benchmark of error-inducing prompts, the company says the error rate at low reasoning effort dropped from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, and that at maximum effort the error rate fell to 4.3% . These figures are especially relevant for business users because low-cost automation is only useful if the outputs do not create constant review burden. But here too, the internal nature of the benchmark means companies should validate Terra against their own documents, policies, and workflows before relying on it at scale.
Availability beyond OpenAI
The launch moved quickly beyond OpenAI’s own surfaces. GitHub announced that GPT-6.1 Terra is generally available in GitHub Copilot for everyday agentic coding and terminal work at a lower cost than GPT-6.1 Sol . GitHub says the model is available to Copilot Pro, Pro+, Max, Business, and Enterprise users, with selection through the model picker across tools including Visual Studio Code, Visual Studio, Copilot CLI, the Copilot cloud agent, GitHub.com, GitHub Mobile, JetBrains IDEs, Xcode, and Eclipse . GitHub also notes that rollout is gradual and that Business and Enterprise administrators can manage access through model policy controls .
AWS also announced general availability of GPT-6.1 Terra on Amazon Bedrock on October 26, 2026 . AWS says customers can call the model directly through Bedrock APIs or configure ChatGPT Work and Codex to use it on Amazon Bedrock . Microsoft Foundry lists GPT-6.1 Terra as a Direct from Azure model with a 2026-10-26 version, hosted on Azure, with text and image input and a 1050k context window . Together, those distribution moves show that Terra is not merely an OpenAI console update; it is being pushed into the developer and enterprise infrastructure where production AI workloads are actually deployed.
Safety and governance
The lower price does not mean lower-risk capability. OpenAI’s system card addendum says the company is treating GPT-6.1 Terra as Critical in cybersecurity and High for biological and chemical capability under its Preparedness Framework . OpenAI says Terra uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, with probes updated to be compatible with the model .
That safety classification is a key part of the current state of the release. It means organizations should not view Terra as a harmless budget model simply because it is cheaper. A model that can use computers, interpret screens, and operate across applications needs access controls, audit logs, data handling rules, and clear limits on what it may do without human approval. For enterprises, Terra’s appeal will depend not only on price, but also on whether administrators can govern it across ChatGPT Work, Codex, Copilot, Bedrock, and Azure.
What Terra means for the AI market
GPT-6.1 Terra is a sign that the model competition is shifting from raw capability to cost-performance. The most advanced model in a family still matters, but many customers now ask a more operational question: which model can complete a given task reliably at the lowest total cost? Terra is designed to answer that question for screen-based workflows, everyday coding, visual reasoning, and structured professional work.
The release also shows how model tiers are becoming more specialized. Astra remains the top-end choice for demanding reasoning, Sol occupies the balance point for complex professional work, and Terra is now the lower-cost model for high-volume tasks that still need meaningful visual and computer-use ability . For developers, the practical decision will be whether Terra is “good enough” to replace Sol or Astra on parts of the workflow. For OpenAI, the strategic goal is clear: make capable AI cheaper to run, easier to deploy, and more attractive for everyday production use.