GPT-6.1 Terra: Price, Benchmarks and What It Can Do

GPT-6.1 Terra Is Here: OpenAI’s Cheaper Model Built to Use Your Computer

OpenAI’s GPT-6.1 Terra is a cheaper, faster model built for computer use, at $1/$5 per million tokens. Here’s what it costs, what OpenAI claims it can do, where you can use it, and why its safety rating stands out.

[email protected]

Oct 27, 2026, 12:42 PM6 min read

Indigo graphic reading GPT-6.1 Terra wants to do your screen work, with an invoice entry form being filled in automatically and a price badge of $1/$5 per million tokens

OpenAI’s newest model isn’t its smartest. That’s the point. GPT-6.1 Terra, released late on Monday, 26 October (around 11:30 PM India time), is built to do the dull computer work that fills a workday: reading a screen, clicking through an app, filling in a form, writing the follow-up email. It does that faster than OpenAI’s bigger models, and for a lot less money.

OpenAI’s launch post on X passed 4.3 million views in under a day. On Hacker News, the announcement drew more than 600 points and 380 comments. Here’s what Terra is, what it costs, what OpenAI claims it can do, and where the catches are.

What is GPT-6.1 Terra?

OpenAI names its current models after celestial bodies. Luna is the small, cheap tier. Terra sits in the middle. Sol is the larger workhorse, and GPT-6 Astra is the company’s most capable frontier model. OpenAI’s own developer docs describe the older GPT-5.6 Terra as roughly the “mini” tier of earlier GPT-5 families, so think of Terra as the mid-priced option.

GPT-6.1 Terra replaces GPT-5.6 Terra in that slot. OpenAI calls it “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing.” Its examples are deliberately unglamorous: drafting proposals, interpreting technical drawings, filling out forms, editing videos, and reading incoming invoices so their details can be typed into accounting software.

On paper, the specs match the bigger models. The API version, gpt-6.1-terra, takes text and images (not audio or video), has a 1,050,000-token context window with up to 128,000 output tokens, and has a knowledge cutoff of 1 June 2026. Developers can set its reasoning effort from “none” up to “max”.

Illustration of a desktop monitor at night showing a generic accounting app: an invoice with the total due highlighted, a form being filled in automatically, and an AI agent task panel noting approval is needed before posting — Illustration

How much does GPT-6.1 Terra cost?

The price is the headline. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that’s 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol. According to OpenAI’s pricing page, the older Terra cost $2 in and $12 out, so the output cut is actually closer to 58%. GPT-6.1 Sol costs $2 in and $10 out, and GPT-6 Astra costs $10 in and $50 out.

There are a few footnotes. Prompts longer than 272,000 input tokens are billed at twice the input rate and 1.5 times the output rate for the whole request. Batch and Flex processing are half price, while Fast mode costs double.

For ChatGPT subscribers, the cost shows up as usage allowance rather than dollars. OpenAI says the same plan allowance in ChatGPT Work and Codex “covers more tasks” with Terra than with GPT-5.6 Terra, and “substantially more” than with GPT-6.1 Sol or GPT-6 Astra.

What OpenAI says it can do

All of the numbers below come from OpenAI and haven’t been independently reproduced unless noted.

Computer use: On Agents’ Last Exam, a benchmark of professional tasks such as turning architectural drawings into a 3D model, OpenAI says Terra at medium effort scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost, and performs on par with GPT-6 Astra at a tenth of the price per token. It says Terra finishes tasks more than three times faster than Astra and GPT-6.1 Sol.
Against Anthropic: At max effort, OpenAI claims Terra is more than three times as token-efficient as Claude Opus 5.5 and a quarter of its price per token, for “~9x lower cost per task.”
Coding: On Terminal-Bench 4.0, Terra at max effort solves 50.3% of tasks, compared with 21.5% for GPT-5.6 Terra. OpenAI says GPT-6.1 Sol and GPT-6 Astra “remain the better choice” for the most complex coding.
Factual errors: On an internal set of prompts designed to trip models up, OpenAI says the error rate at low effort fell from 53.6% to 16.7%, and at max effort fell 18 points to 4.3%.

The few outside numbers so far are encouraging but limited. Independent benchmarker Artificial Analysis scored GPT-6.1 Terra at max effort at 48 on its Intelligence Index (v4.3.2), against a median of 28 for models in its price tier. It measured about $0.40 per index task and 91 output tokens per second, but also flagged it as “somewhat verbose” and recorded a very long wait before the first token at max effort.

Infographic titled GPT-6.1 Terra at a glance: API prices per million tokens for GPT-6 Luna, GPT-6.1 Terra, GPT-5.6 Terra, GPT-6.1 Sol and GPT-6 Astra; OpenAI-reported benchmark claims; availability in ChatGPT Work, Codex and the API; and its Critical cybersecurity rating

Where you can use it, and where you can’t

In ChatGPT, Terra isn’t in the regular chat window. OpenAI’s help documentation says it’s available in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans, chosen from the model list in the desktop app or ChatGPT Work on the web, and that “the model isn’t available in Chat.” Enterprise access depends on admin controls, and browser and computer use each have their own setup requirements.

Developers get it through the API, where tool calling needs the Responses API. Microsoft also made it generally available in Microsoft Foundry on launch day, with data-zone deployments in the US, EU and Asia-Pacific.

That makes Terra a natural fit for the always-on assistants now arriving from every big lab, which we covered when Claude Duet joined OpenAI’s Dots and Meta’s Muse. Agents that click through software all day burn a lot of tokens, so a cheaper, faster mid-tier model matters more there than one more point on a reasoning benchmark.

The safety fine print

This is the part worth reading twice. In its system card addendum, OpenAI says it is treating GPT-6.1 Terra as “Critical” in cybersecurity under its Preparedness Framework, the top level. That framework defines Critical as a model that can “identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention,” or run novel end-to-end attack strategies against hardened targets. Terra is also rated “High” for biological and chemical capability. It ships with the same safeguards as GPT-6.1 Sol and GPT-6 Astra.

In other words, a model at the mid-tier price point is rated at the top cyber-risk level OpenAI uses. On ExploitBench, which tests turning known vulnerabilities into working exploits, Terra scored 97.10%, though OpenAI itself warns the benchmark may be inflated by contamination.

Illustration of a laptop browser on a supplier checkout page where hidden text tries to instruct the AI agent to change a payout account; the agent's task log marks it as ignored and asks the user to approve or stop a payment — Illustration

Computer use brings its own risk: prompt injection, where text hidden in a web page or document tries to hijack the agent. OpenAI says Terra is “similar to” Astra and GPT-6.1 Sol, which it calls “substantially robust” to indirect injections, and that it is still updating those evaluations. In a simulation of internal Codex traffic across 48,690 tasks, Terra received 22 serious misalignment flags (0.045%), fewer than any of the four other models compared. But at lower severity levels, it was flagged more often than Sol and Astra for behaviours including circumventing restrictions, destructive actions and inbound prompt injection. If you plan to let it loose on real accounts, the agent guardrail tools vendors launched last week are worth a look.

Should you switch?

Not everyone is sold. Some Hacker News commenters called Terra a confusing middle child between Luna and Sol, and others said they’d rather OpenAI shipped a model to beat Claude Opus 5.5 outright. That’s a fair point. Terra isn’t trying to top the leaderboards. For hard coding problems, OpenAI itself points you to Sol or Astra, and open-weight rivals like Moonshot’s Kimi K3.1 Pro are competing hard on price.

Where Terra makes sense is volume. If you run lots of document processing, form filling, screen reading or customer-facing drafting, a model that OpenAI says matches Astra on agent tasks at a tenth of the token price is worth testing against your current setup. Run your own tasks, watch the token counts at high reasoning effort, and keep a human approval step for anything that spends money or sends messages.

It’s also been a busy week for OpenAI. Terra landed on the same day the company said an internal model had refuted the Exponential Time Hypothesis. One release shows off the frontier; the other is the one far more people will actually use.

Related, 28 October: a day later, Google DeepMind showed AI agents of a different kind: Genie 4, a world model that builds explorable, talking simulations for robots and creators.

Sources

Automation Explainer