GPT-6.1 Terra: OpenAI's $1 Computer-Use Model, 3x Faster Than Astra

The short version: OpenAI released GPT-6.1 Terra on October 26, 2026, a fast, efficient model built to do the computer work that fills a workday: reading what's on screen, clicking through apps, filling in forms, drafting proposals, and editing video. Its standard API price is $1 per million input tokens and $5 per million output tokens — half the price of GPT-6.1 Sol and, on output, well under half the $12 GPT-5.6 Terra charged. OpenAI says it matches the flagship GPT-6 Astra on computer-use tasks at one-tenth the price per token and finishes those tasks more than three times faster. The catch sits in the safety section of the same announcement: Terra meets OpenAI's Critical threshold for cybersecurity capability, the highest tier in its safety framework.

Terra fills the gap OpenAI left open a month ago. When GPT-6.1 Sol and GPT-6 Luna arrived in late September, the Terra tier stayed a generation behind on GPT-5.6. This release closes that gap — and the price cut is the point: Terra is the model you can afford to leave running all day.

What GPT-6.1 Terra actually does

Terra is a computer-use model. That means it reads a screen, understands what it is looking at, and takes action across applications — the loop of seeing, clicking, typing, and checking that makes up most office work. OpenAI's own examples are deliberately unglamorous: drafting proposals, interpreting technical drawings, filling out forms, editing video, and reading incoming invoices and entering their details into accounting software.

Two things in the announcement are worth noting for what they rule out. First, Terra takes text and image input and produces text — native audio and video input are not part of the model, according to OpenAI's developer documentation. Second, OpenAI is explicit about what Terra is not for: "For in-depth research or complex coding, choose GPT-6.1 Sol or GPT-6 Astra." Terra is the workday model; Sol and Astra keep the hardest jobs.

The model also brings clearer, more natural writing, according to OpenAI — a deliberate pitch at the communication half of office work, not just the clicking half. It ships with a 1,050,000-token context window, a 128,000-token maximum output, and a knowledge cutoff of June 1, 2026, per OpenAI's developer documentation. A million-token window matters for computer use in particular: an agent holding a long document, a set of screenshots, and a running transcript in context at once needs far fewer round trips.

What it costs — and the fine print

The headline number is simple: $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. OpenAI describes that as 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. The rate card tells a slightly better story on output than the headline suggests: GPT-5.6 Terra charged $12 per million output tokens, so Terra's $5 is about 58% lower on that line, not 50%. Input is exactly half — $1 against $2 for both the predecessor and Sol.

The fine print is where the real bill lives, and there are four items worth knowing before you build on it:

Long prompts cost double. OpenAI's pricing page charges higher rates for prompts above 272,000 input tokens. For Terra that means $2 per million input and $10 per million output once you cross that line — the same as Sol's standard rate. A million-token context window is only cheap if your prompts stay under the threshold.

Cache writes are not free. Cached input is nearly free at $0.05 per million tokens, but writing to the cache costs $1.25 per million. That is still half of GPT-5.6 Terra's $2.50 cache-write rate, per Artificial Analysis, and it rewards repeated context — the same reference material, form layouts, or catalog data reused across calls — rather than one-off prompts.

Batch and Flex halve the price again. OpenAI's pricing page lists Batch and Flex processing for Terra at $0.50 per million input and $2.50 per million output. If your workload can wait or flex, the effective price drops to a quarter of what GPT-5.6 Terra charged at standard rates.

Inside ChatGPT, there is no per-token bill at all. Terra is rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and OpenAI says the same usage allowance covers more tasks with Terra than with GPT-5.6 Terra — and "substantially more" than with Sol or Astra. Plus starts at $20 a month, so the cheapest way to try Terra is a Plus subscription, not the API. The tradeoff is control: in ChatGPT you get the model, not the knobs.

The benchmarks OpenAI is showing

Every number in this section is OpenAI's own, from its launch announcement. That matters, because the company chose the tests and ran them — treat the figures as a claim about relative performance, not an independent measurement. With that caveat, the pattern is consistent: Terra closes most of the gap to the flagship on computer-use tasks, at a fraction of the price.

Agents' Last Exam is the headline test — a benchmark of professional tasks that includes turning architectural drawings into a 3D model and reviewing an animated film shot by shot to build a time-coded log for a restoration team. At medium reasoning effort, Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. It performs on par with GPT-6 Astra at one-tenth the price per token. At maximum effort, OpenAI says Terra is over three times as token-efficient as Claude Opus 5.5 and one-fourth its price per token, producing a roughly 9x lower cost per task. And across the board, Terra completes tasks over three times faster than both Astra and Sol.

MMMU-Pro, a visual-reasoning benchmark, shows Terra outperforming GPT-5.6 Terra at every reasoning setting and nearly matching Astra and Sol at much lower cost. OSWorld 2.1, which tests agents using desktop apps for multistep tasks such as building a 3D model from a technical drawing, has Terra surpassing GPT-5.6 Terra's best score at medium effort for a fraction of the cost per task.

The factuality numbers are the most striking in the announcement. On OpenAI's 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 Terra — a drop of nearly 37 percentage points. At maximum effort the error rate fell to 4.3%, down 18 points, nearly matching Astra and Sol.

On coding, Terminal-Bench 4.0 has Terra solving 50.3% of tasks at maximum reasoning effort, more than double GPT-5.6 Terra's published 21.5%. And on Terminal-Bench-Science 0.1, which tests agents using code for scientific research tasks, Terra surpasses GPT-6.1 Sol at lower cost at maximum reasoning, coming within five points of Claude Opus 5.5 at less than one-fifth the estimated cost. OpenAI still points users to Sol or Astra for the most complex coding — Terra's win here is on breadth and price, not on the hardest problems.

What independent testing says

Artificial Analysis published its own evaluation of Terra on launch day, and it is broadly supportive — with the kind of nuance vendor announcements skip. Pricing per token is half or less than GPT-5.6 Terra, the firm confirms: $1/$5 against $2/$12, cache write halved to $1.25, cached input a quarter of the old price at $0.05.

On its Intelligence Index, Terra gains 6 points over GPT-5.6 Terra at maximum effort while costing less than a third as much per task — $0.40 against $1.40. At low and medium effort the gain is larger, between 7 and 13 points, and those effort levels sit on the cost-efficiency frontier. At maximum effort Terra scores 4 points below GPT-6.1 Sol and 10 points above GPT-6 Luna. On the firm's Coding Agent Index v2.0, Terra scores 48, six points below Sol's 54 and 17 above Luna's 31.

The hallucination result is the one to remember. On Artificial Analysis's private knowledge benchmark, AA-Omniscience, Terra guesses 38% of the time when it does not know an answer — down from 88% for GPT-5.6 Terra. That 50-point drop comes with a 2-point increase in accuracy, meaning the model answers more questions correctly while guessing far less often. For a computer-use agent, a model that says "I don't know" instead of clicking the wrong button is worth more than a few extra benchmark points.

Terra vs the field

Two rows in that table deserve a second look. The first is Luna: at $0.10/$0.50 it is ten times cheaper than Terra on input — but Luna is not a computer-use model. Terra's competition is not the cheapest model in OpenAI's lineup; it is the cheapest model that can actually drive software. The second is Opus 5.5. OpenAI benchmarks Terra against Anthropic's flagship throughout its announcement, and the comparison is pointed: Terra scores slightly below Opus on Agents' Last Exam at maximum effort, but at roughly a ninth of the cost per task. OpenAI is not claiming Terra is smarter than Opus. It is claiming the gap is small enough that the price difference decides it — the same argument that has driven the last four months of price cuts across the industry, which we covered in our analysis of OpenAI and Anthropic cutting AI prices by up to 80% in four months.

The safety flag worth reading twice

Buried at the end of OpenAI's announcement is the sentence that matters most for anyone planning to let Terra loose on a real machine: "GPT-6.1 Terra meets the Critical threshold in cybersecurity under our Preparedness Framework, and maintains the protections developed for GPT-6 Astra."

"Critical" is the highest capability tier in OpenAI's Preparedness Framework — the level at which a model's cybersecurity abilities are considered advanced enough to require the strongest safeguards. It is a rating of what the model can do, not a report of an incident, and OpenAI says Terra carries the protections developed for Astra. The announcement also says Terra is more likely than GPT-5.6 Terra to stay within scope and honestly disclose limitations and blockers, and that it showed strong resistance in prompt-injection tests — tests where malicious instructions try to divert the model from the user's task. More detail is in the published system card.

The practical read is straightforward: a model that can operate a computer can also do damage on one, by mistake or by manipulation. Terra's own numbers point the right way — it guesses far less and discloses limits more honestly than its predecessor — but the Critical rating is a reminder that the safeguards are doing real work here. If you deploy Terra in an agent, sandbox it, scope its permissions, and read the system card before you hand it credentials.

Where to get it, and who should switch

Terra is available now in three places: ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users (Business per OpenAI's release notes); the OpenAI API as gpt-6.1-terra; and, according to launch coverage, Amazon Bedrock and GitHub Copilot. Codex CLI users need version 0.159.0 or later, per launch coverage. Availability in ChatGPT depends on plan, region, and rollout progress, so some users will see it a few days after others.

Who should switch is a short list. If you run agents that do repetitive computer work — invoice entry, form filling, proposal drafting, video editing timelines — Terra is the cheapest OpenAI model that can actually do the job, and the 3x speed gain over Sol and Astra matters as much as the price: agents that finish faster cost less twice over. If you are on GPT-5.6 Terra, the switch is a straight upgrade on every published number. If your workload is the hardest coding or deep research, stay on Sol or Astra — OpenAI says so itself.

One more consideration: Terra's value depends on how you use the context window. The 1,050,000-token window is a genuine asset for agents that juggle documents and screenshots, but prompts above 272,000 tokens are charged at double the standard rate. If your agent's prompts routinely cross that line, Terra's price advantage over Sol narrows to nothing. Keep prompts lean, cache what repeats, and use Batch or Flex where the work can wait — that is where the $1 price actually holds.

Pros and cons

Pros: Half the price of GPT-5.6 Terra and GPT-6.1 Sol on every line of the rate card; over 3x faster than Astra and Sol on agent tasks; matches Astra on Agents' Last Exam at one-tenth the price per token; guessing rate cut from 88% to 38% in independent testing; 1.05M-token context window; available everywhere at once — ChatGPT, Codex, API, Bedrock, and Copilot.

Cons: Not the pick for the hardest coding or research — Sol and Astra keep those jobs; every headline benchmark is OpenAI's own, with independent confirmation so far limited to Artificial Analysis; meets the Critical cybersecurity threshold, so deployment needs real sandboxing and permission scoping; long prompts above 272,000 tokens are charged at double the rate; text and image input only, no native audio or video.

FAQ

What is GPT-6.1 Terra?
It is OpenAI's mid-tier model for computer use, launched October 26, 2026. It reads screens, takes action across apps, and writes clearly — built for the communication and computer-based tasks that fill a workday, from drafting proposals to filling in forms and editing video. It replaces GPT-5.6 Terra.

How much does GPT-6.1 Terra cost?
Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. Cache writes cost $1.25 per million. Batch and Flex processing run at half those rates. Prompts above 272,000 input tokens are charged at double the standard rate. In ChatGPT, Terra is included in Plus, Pro, Business, Enterprise, and Edu plans with no separate per-token charge.

Is GPT-6.1 Terra better than GPT-6.1 Sol?
Not in general — they are built for different jobs. Terra is faster and cheaper for computer use and workday tasks, and it beats Sol on Terminal-Bench-Science 0.1 at lower cost, according to OpenAI. But for the most complex coding and in-depth research, OpenAI still recommends Sol or the flagship Astra. Terra's edge is price and speed on everyday agent work, not peak capability.

Is GPT-6.1 Terra available on the free ChatGPT plan?
OpenAI has announced Terra for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. Free-tier access has not been announced. The API is open to developers as gpt-6.1-terra.

What does the Critical cybersecurity rating mean?
It means Terra's cybersecurity capabilities meet the highest tier in OpenAI's Preparedness Framework — the level that triggers the strongest safeguards. OpenAI says Terra maintains the protections developed for GPT-6 Astra, and that it showed strong resistance in prompt-injection tests. The full detail is in the published system card.

Key takeaways

• OpenAI released GPT-6.1 Terra on October 26, 2026, at $1 per million input tokens and $5 per million output tokens — half the price of GPT-6.1 Sol and GPT-5.6 Terra.

• On Agents' Last Exam it scores 15 points higher than GPT-5.6 Terra at medium effort, matches the flagship GPT-6 Astra, and finishes tasks over three times faster than Astra and Sol.

• Independent testing from Artificial Analysis confirms the value: a 6-point Intelligence Index gain over the predecessor at less than a third of the cost per task, and a guessing rate cut from 88% to 38%.

• The fine print matters: prompts above 272,000 tokens cost double, cache writes are $1.25 per million, and Batch/Flex processing halves the standard rates.

• Terra meets OpenAI's Critical cybersecurity threshold — the highest tier in its safety framework — so anyone deploying it as an agent should sandbox it, scope its permissions, and read the system card first.

Terra is not OpenAI's smartest model, and the company does not pretend otherwise. It is the cheapest model that can actually drive a computer all day — and as of October 26, 2026, that is a new rung on the ladder, priced to be used rather than admired. For the full context on the price war that made a $1 computer-use model possible, see our analysis of OpenAI's next-generation reasoning models — including GPT-6.1 Sol and the Decisions API — and the GPT-6 launch on Google Cloud.

Sources: OpenAI — Introducing GPT-6.1 Terra · OpenAI API pricing · Artificial Analysis — GPT-6.1 Terra pushes the cost efficiency frontier · BleepingComputer — OpenAI launches GPT-6.1 Terra · Claude pricing