OpenAI's GPT-6.1 Terra: Cheaper Computer Use for Real Work

OpenAI's GPT-6.1 Terra: Cheaper Computer Use for Real Work

On October 26, 2026, OpenAI released GPT-6.1 Terra, a model aimed at the work that fills most business days: reading screens, filling forms, drafting, and moving information between tools. The pitch is not a new chatbot personality. It is speed, computer use, and a lower API bill.

If you run a real business, that combination matters more than another benchmark chart. The question is whether the unit economics and the error rates hold up once you put the model on invoice entry, CRM updates, or desktop workflows.

What OpenAI shipped

OpenAI describes Terra as a fast, efficient model that combines visual reasoning and precise computer use with clearer writing. It positions the release as a step up from GPT-5.6 Terra for the communication and computer-based tasks that fill the workday.

Availability is clear. Starting October 26, GPT-6.1 Terra is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can call it through the API as gpt-6.1-terra. The Verge reported the same day with a Posted Oct 26, 2026 timestamp.

The price cut

The pricing is the first number operators should write down.

OpenAI states that Terra's standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. Standard API pricing is:

• $1 per million input tokens
• $0.05 per million cached input tokens
• $5 per million output tokens

OpenAI also says that in ChatGPT Work and Codex, the same usage allowance covers more tasks with Terra than with GPT-5.6 Terra, and substantially more than with GPT-6.1 Sol or GPT-6 Astra.

That is OpenAI's claim. It is useful as a planning input, not as a guarantee that your workflow will cost half as much. Your cost depends on how many tokens your agent burns per task, how often it retries, and how much human review you still need.

Computer use and the invoice example

OpenAI is explicit about the kind of work it wants Terra to automate. The company says lower prices make it practical for developers to automate high-volume work, "like reading incoming invoices and entering their details into accounting software."

That example is not abstract. Invoice capture, data entry, and desktop form filling are still expensive when a person does them all day. They are also the jobs where a cheaper computer-use model either saves real money or creates expensive cleanup if it invents line items.

OpenAI reports that at medium reasoning effort on Agents' Last Exam, Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. It also states that Terra completes tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol, and claims about 9x lower cost per task versus Claude Opus 5.5 at max effort, based on token efficiency plus price. Those are OpenAI's comparisons on its chosen benchmarks. Treat them as vendor-reported, not as independent audits of your stack.

Error rates and coding scores

Price without reliability is a trap. OpenAI published internal numbers on factual errors.

On an internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. At maximum effort on that same internal benchmark, the error rate fell by 18 percentage points to 4.3%.

Those figures come from OpenAI. They are directionally useful: OpenAI is saying low-effort mode is much less wrong than the previous Terra, and max effort lands near its higher-tier models on that internal test. They are not a warranty for your invoices, customer emails, or tax forms.

On coding, OpenAI says that on Terminal-Bench 4.0 at maximum reasoning effort, GPT-6.1 Terra solves 50.3% of tasks, more than double GPT-5.6 Terra's published score of 21.5%. OpenAI still says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks.

What this is not

This release does not mean every desktop agent is production-ready. OpenAI's own framing is about a better cost and capability tradeoff for computer-based work. It does not publish customer ROI studies in the launch post. It does not claim zero hallucination. And a system card addendum was published October 26, 2026 for operators who need the safety detail before they wire Terra into a paid workflow.

If your process cannot tolerate a wrong amount on a bill, you still need a check step. Cheaper tokens do not remove that requirement. They change how often you can afford to run the check.

How to evaluate it this week

If you already pay for ChatGPT Work or Codex on a Plus, Pro, Enterprise, or Edu plan, you can try Terra without a new contract. If you build through the API, run a side-by-side on one workflow you already measure:

1. Pick a task with a clear unit cost today, such as invoice entry or form fill.
2. Log tokens, wall time, and error rate against a human or your current model.
3. Keep a human approval gate on money movement, deletions, and customer-facing sends.
4. Recalculate cost per successful task, not cost per token. Retries and cleanup count.

That is the operator test. Benchmarks tell you the model improved. Your ledger tells you whether it belongs in the stack.

What it means for operators (Jaden's take)

I care about tools that change the cost of work I already do. Terra is interesting because OpenAI put the invoice example in the launch post and cut the list price in half versus the prior Terra and Sol tiers. That is a concrete reason to re-check agent workflows, not a reason to rewrite your whole stack overnight.

My rule is simple. Use the cheaper computer-use model where the output is easy to verify. Keep expensive frontier models, or a person, where a wrong answer is expensive. If Terra can pull invoice fields into accounting software with a review step, that is real leverage. If it invents numbers and you only notice at month end, that is not leverage. That is debt.

Re-run the unit economics this month. Keep the human gate on money. Ignore the hype charts until your own error log agrees.