OpenAI's GPT-6.1 Terra Halves API Prices for Enterprise Computer-Use Agents

OpenAI released GPT-6.1 Terra on Monday, October 26, a model built for high-volume office work such as reading invoices, filling out forms and operating desktop software. Its standard API price is half that of GPT-5.6 Terra and GPT-6.1 Sol, according to PYMNTS. For enterprises that have shelved computer-use agents because the per-task cost did not add up, the launch changes the math, and it makes cost governance more urgent, not less.

What is GPT-6.1 Terra?

Terra sits in the middle of OpenAI's current lineup. Microsoft's Foundry team describes it as a fast, efficient model for everyday professional tasks that complements GPT-6 Astra, which handles open-ended, multi-step planning. Terra's job is the repetitive, screen-based work: matching supplier photos to catalog entries, reading technical drawings and floor plans, reviewing campaign assets, and moving through browsers and desktop applications.

It accepts text and images, supports up to 1 million input tokens, and lets developers dial reasoning effort down to "none" for latency-sensitive calls, Microsoft said. It can also send progress updates to users during long tasks if the orchestration layer supports it.

OpenAI made the model available the same day in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and to developers through its API, PYMNTS reported. Microsoft made it generally available in Foundry on October 26, with Data Zone deployments in the US, EU and Asia-Pacific for customers with data residency requirements.

How much cheaper is it, and does it perform?

Standard API pricing is $1 per million input tokens, $0.05 for cached input and $5 per million output tokens, Business Today reported. Microsoft's Foundry price sheet matches those figures for short-context Global Standard deployments, with higher rates for long contexts and for regional Data Zone deployments, where EU and Asia-Pacific prices run about 20% above the global rate.

On performance, OpenAI's own evaluations claim Terra scores 15 percentage points higher than GPT-5.6 Terra on the Agents' Last Exam benchmark at medium reasoning effort, at roughly a third of the cost per task, and 50.3% on Terminal-Bench 4.0 versus 21.5% for its predecessor, according to Business Today. These are vendor numbers and have not been independently verified. OpenAI framed the release as offering "a new balance of speed, capability and cost."

The model also arrives as OpenAI's enterprise business grows quickly. PYMNTS, citing Bloomberg reporting from earlier in October, said OpenAI's annualized revenue is expected to reach at least $70 billion by year-end.

Why does a cheaper computer-use model matter for enterprises?

Computer-use agents promise to automate work in systems that have no clean API, which describes a large share of enterprise back-office software. The barrier has been cost per task, since an agent that reads screens and clicks through applications consumes many tokens per step. Halving the price of a capable model brings more of those workflows within reach of a business case.

That does not mean every workflow should move to the new model. Pushkar Rege, Vice President and Co-Founder of the Business Intelligence Program at Mastercard, made that point at The AI Enterprise Conference 2026. "You don't need premium models for everything that you're doing," he said. His fellow panelists argued that teams should track AI cost per outcome rather than cost per token, and match model size to the task. A mid-tier model such as Terra fits that approach, but only if teams measure what each task actually costs end to end.

Lower prices can also hide a trap. Verisk Global Chief Data Officer Louis Dimodugno warned at the same conference that token prices are changing almost daily as model providers compete for market share, and could rise once those companies go public. "The financial piece is the piece that's always going to get the most attention," he said. His advice, covered in our look at self-funding AI strategies, is to build cost visibility before usage scales.

What are the risks of agents that operate the desktop?

An agent that can click, type and submit across applications can also do the wrong thing quickly. Business Today noted that Terra's computer-use functions depend on the tools and permissions a deployment grants, and advised users to review important actions rather than assume every task will be completed correctly.

That matches what security practitioners have told AI Connect audiences. Rubrik's Cal Al-Dhubaib described real cases of AI agent overreach, including an agent that published an internal report to a public site after hitting an error. Before giving a cheaper agent access to finance or procurement screens, enterprises should define which actions require human approval, scope credentials to the task, and log every step.

What to watch next

The near-term test is whether Terra's benchmark gains hold up on real enterprise screens, where legacy interfaces, pop-ups and inconsistent layouts trip up agents. Expect rival vendors to respond on price, and expect finance teams to ask a sharper question: not what a model costs per million tokens, but what a completed invoice, form or catalog update costs once retries, reviews and errors are counted.

Frequently asked questions

What is GPT-6.1 Terra?

It is an OpenAI model released October 26, 2026, designed for fast, lower-cost professional tasks such as computer use, visual reasoning, coding and writing.

How much does GPT-6.1 Terra cost?

Standard API pricing is $1 per million input tokens, $0.05 for cached input and $5 per million output tokens, with higher rates for long contexts and regional deployments in Microsoft Foundry.

Where is GPT-6.1 Terra available?

It is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, through the OpenAI API, and in Microsoft Foundry, including US, EU and Asia-Pacific Data Zone deployments.

Filed under: Agentic AI, AI Strategy & ROI