OpenAI's GPT-6.1 Terra, launched on 26 October 2026, is a half-price model built for business automation: it costs 50% less than GPT-5.6 Terra and GPT-6.1 Sol, and it targets the everyday work that eats up staff hours — drafting proposals, interpreting drawings, filling forms, editing video, reading invoices and entering data into accounting software.

It is available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API as gpt-6.1-terra. Standard API pricing is $1 per 1 million input tokens, $0.05 per 1 million cached input tokens and $5 per 1 million output tokens.

GPT-6.1 Terra: Half-Price AI for Business Automation

What GPT-6.1 Terra actually is

GPT-6.1 Terra is OpenAI's lower-cost, high-volume model in the GPT-6.1 family, built for agent workflows: multi-step tasks where the model reads documents, uses tools, checks its own output and carries the job through to completion. Its huge context, long output and adjustable effort make it a workhorse rather than a showpiece.

The headline specifications

• Context window of 1,050,000 tokens — enough to hold very large documents and whole sets of records in a single run.
• Up to 128,000 output tokens per run, so long drafts and processed batches are not cut short.
• Knowledge cutoff of 1 June 2026.
• Six reasoning effort levels, so you can dial the model up for a tricky job or down for a routine one.

That combination — huge context, long output and adjustable effort — is what makes it a workhorse rather than a showpiece.

The price: half the cost, in plain numbers

At $1 per 1 million input tokens, $0.05 cached and $5 output, GPT-6.1 Terra costs 50% less than GPT-5.6 Terra and GPT-6.1 Sol. That halving matters because agent workloads consume far more tokens than a single chat reply: the agent reads, reasons, acts and re-checks, so when every step costs half as much, the saving compounds across every task, every day.

The pricing details that change the bill

• Prompts over 272,000 input tokens are billed at 2x the input and cached-input rates and 1.5x the output rate.
• Batch and Flex processing cut prices by 50% again, making overnight or non-urgent bulk work cheaper still.
• Fast processing doubles the prices, for work where speed matters more than cost.

Route routine bulk work through Batch or Flex where timing allows, keep Fast for genuinely urgent jobs, and watch the long-prompt surcharge on very large document sets. That is the same discipline behind any serious AI automation project: price the whole workflow, not the single call. Any wider cost projection beyond these published rates should be treated as an estimate.

What the benchmarks say — attributed to OpenAI's own evaluations

Benchmark claims for a new model should always carry their source, and here the source is OpenAI's own evaluations. On Agents' Last Exam, GPT-6.1 Terra at medium effort scored 15 points above GPT-5.6 Terra, at roughly one-third of the cost. On Terminal-Bench 4.0, it scored 50.3% at maximum effort, versus 21.5% for GPT-5.6 Terra. Both figures point the same way: materially stronger agent performance at a materially lower price.

What to measure instead of benchmarks

Benchmarks are a starting point, not a verdict — they measure test tasks, not your invoices or your clients' formats. The honest reading is that GPT-6.1 Terra has earned a proper pilot on your own tasks. Measure success rate on your own routine work, checked by a human reviewer; tokens and cost per completed task, including retries; how often staff must intervene; and whether the output respects your formats and approval steps. Teams building AI agents for production should run that pilot with fixed tasks and recorded success criteria.

Where GPT-6.1 Terra fits: everyday business automation

Proposals, forms and documents

Drafting proposals from a brief, filling in standard forms and turning rough notes into client-ready documents are ideal first pilots: high-volume, repetitive and easy to check, with a human reviewer able to approve or correct output quickly. Businesses exploring SaaS development or custom software development often find document workflows are the fastest automation win.

Invoices, data entry and accounting

Reading invoices and entering data into accounting software is the classic automation candidate: structured, repetitive and error-prone by hand. With a 1,050,000-token context, an agent can work through large batches in a single run, and half-price tokens make that batch work affordable. Keep a human approval step for payments and exceptions, and the risk stays controlled while the hours saved grow.

Drawings, video and customer-facing work

Interpreting drawings and editing video matter for construction, engineering, manufacturing and media businesses, where turning drawings into summaries or cutting routine video eats hours. The same cheaper agent capacity can power customer-facing work: a chatbot, a WhatsApp bot, or a voice-led system such as the AI calling agent. For hospitality, the same logic applies inside restaurant ERP.

What to watch before you commit

Half price is a strong headline, but three details deserve attention. First, the long-prompt surcharge: prompts over 272,000 input tokens bill at 2x input and cached-input rates and 1.5x output, so test whether splitting a job or trimming context keeps you under that threshold. Second, the speed tiers pull in opposite directions — Batch and Flex halve the price again, Fast doubles it — so routing is part of the design. Third, six reasoning effort levels mean the same task can cost very different amounts: use low effort for routine work and reserve high effort for jobs where accuracy pays.

Availability, access and governance

GPT-6.1 Terra is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and via the API as gpt-6.1-terra. Confirm which of your plans includes it before promising a rollout date, and check where your data goes via the API, what retention applies, and whether that fits your client contracts. Businesses in the United Kingdom, the UAE and Pakistan face different regulatory expectations, so do not import another company's governance answer wholesale.

Coverage of the launch, including the enterprise-automation angle, appeared in PYMNTS and Chosun Biz.

How to pilot GPT-6.1 Terra in your business

Pick one workflow, run it on GPT-6.1 Terra and on your current approach, and compare like with like.

A four-week pilot plan

• Week 1: pick one routine workflow — invoice entry, proposal drafting or form filling — and record the current time, cost and error rate.
• Week 2: run the same tasks through a GPT-6.1 Terra agent with a human reviewer on every output.
• Week 3: measure tokens per completed task, reviewer effort and exception rate, and test Batch or Flex routing for non-urgent batches.
• Week 4: decide on evidence — scale it, adjust it, or keep the old process for that workflow and try the next one.

If you need help designing the pilot, read our guides on chatbots versus AI agents, how to build an AI voice agent and AI calling agent costs, and apply the same logic to ecommerce development and CRM development projects.

Frequently asked questions

When was GPT-6.1 Terra launched, and where is it available?

GPT-6.1 Terra launched on 26 October 2026. It is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API as gpt-6.1-terra.

How much does the GPT-6.1 Terra API cost?

Standard API pricing is $1 per 1 million input tokens, $0.05 per 1 million cached input tokens and $5 per 1 million output tokens — 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Prompts over 272,000 input tokens are billed at 2x input and cached-input rates and 1.5x output; Batch and Flex cut prices by 50%, while Fast doubles them.

What is GPT-6.1 Terra best used for?

It is suited to everyday business automation: drafting proposals, interpreting drawings, filling forms, editing video, reading invoices and entering data into accounting software. Its huge context and six effort levels suit high-volume, multi-step agent workflows.

Should my business switch to GPT-6.1 Terra now?

Pilot it first. On OpenAI's own evaluations it scored 15 points above GPT-5.6 Terra on Agents' Last Exam at medium effort for roughly one-third of the cost, and 50.3% versus 21.5% on Terminal-Bench 4.0 at maximum effort — but benchmarks are not proof for your tasks. Run a short, recorded pilot on your own workflows, measure cost per completed task, and scale only what the evidence supports.