OpenAI launches GPT-6.1 Terra with computer use at half the API price

What happened

OpenAI released GPT-6.1 Terra on 26 October 2026. According to the company's launch post, the model is available to all Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can reach it through the OpenAI API under the model name gpt-6.1-terra. OpenAI describes it as a fast, efficient model for visual reasoning, precise computer use and clear writing.

The pricing is the headline for buyers of automation. OpenAI states standard API pricing of $1 per million input tokens and $5 per million output tokens, along with $0.05 per million cached input tokens. The company says these rates sit 50% below GPT-5.6 Terra and GPT-6.1 Sol, the two models it replaces or sits beside in its range. OpenAI gives invoice entry into accounting software as one example of the computer-use work the model is meant for. Bleepingcomputer reported the same launch and figures on 26 October 2026, citing OpenAI's blog post.

Why it matters for businesses that automate work

Automated workers that operate software through screens do their job in many small steps. Each screenshot read, form field and click consumes tokens, so the per-token price is a direct input to the running cost of that work. A lower price matters most where a process runs thousands of times a month.

OpenAI's own benchmark claims point the same way, though they come from the company itself. On Agents' Last Exam, OpenAI says Terra scores 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort, at about one-third of the cost. On Terminal-Bench 4.0 at maximum effort, the company reports a 50.3% task completion rate, against 21.5% for GPT-5.6 Terra. OpenAI also compares cost per task with Anthropic's Claude Opus 5.5 and says Terra comes out about nine times lower at maximum effort. Independent third-party benchmarks had not been published at the time of the launch, according to FinanceFeeds.

Key figures from the launch

What changes in practice

For a business that already uses an AI tool, the first change is cost per run, not capability. An owner or operations head can ask the team to test the model on a fixed batch of work, such as a month of supplier invoices, and compare the result with the current method. The test should measure how many entries were correct and how long a person spent checking them, not only the token bill.

Access depends on the plan. The launch post lists Plus, Pro, Enterprise and Edu. OpenAI's own release notes for 26 to 30 October 2026 also list Business plans. The two pages do not match on that point, so a business should check the model list in its own workspace before planning around it.

Developers have one extra requirement. OpenAI's model documentation says that the Codex CLI needs version 0.159.0 or later to run GPT-6.1 Terra. Businesses with an internal developer should confirm the version before starting a pilot.

Limits and risks

A computer-use model works by acting on screens, and a wrong entry in a ledger can be costly to trace. OpenAI's internal test reports an error rate of 16.7% at low reasoning effort, down from 53.6% for GPT-5.6 Terra, which means errors are still possible at the lowest setting. OpenAI still recommends its Sol or Astra models for complex coding and multi-step problem solving. Terra is aimed at everyday writing, visual and screen work.

Automation is not worth the set-up for every process. A business that receives a handful of invoices a month may find that a person entering them is faster than building and checking an agent. If a change to the software work also alters anyone's role, hours or pay, the notice, consultation and severance rules of the country concerned apply, and the change should be agreed with the employee or checked with a lawyer.

What to watch next

Three things will show whether the launch holds up outside OpenAI's own tests. The first is independent benchmark results, which had not been published at launch. The second is whether the plan list is clarified, since the launch post and the release notes differ on Business access. The third is whether OpenAI publishes cost per completed task for typical office workflows, rather than only price per token. Until then, the price figures are firm as OpenAI's stated rates, and the performance claims remain the company's own.