AI Agent UpdateOpenAI launches GPT-6.1 Terra at half the API price for enterprise agents
OpenAI released GPT-6.1 Terra at half the API price of its siblings — $1 per million input tokens and $5 per million output tokens — aimed at the everyday computer work of office automation.
The News
OpenAI released GPT-6.1 Terra on October 26, a new model aimed squarely at enterprise automation — and priced to make that automation cheap. Standard API pricing is $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Terra is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers as gpt-6.1-terra. Amazon Web Services made it available on Bedrock the same day.
The pitch is not raw intelligence but workday usefulness. OpenAI positions Terra for computer-use tasks — forms, invoices, proposals, technical drawings, video editing — the repetitive office work where agents either save real money or do not get used at all. Its context window, the amount of material the model can hold in mind at once, is 1.05 million tokens. OpenAI says Terra is 15 percentage points better than GPT-5.6 Terra on Agents' Last Exam, a benchmark designed for agent tasks, at about a third of the per-task cost on that benchmark — a different measure from the 50% token-price cut.
The price cut is the story, and it fits a pattern. Frontier labs are now competing on the cost of running agents, not just on how smart those agents are, because enterprise buyers do the arithmetic: an agent that costs twice as much must be twice as useful, or it loses. Halving the price while claiming better agent performance is OpenAI's attempt to win that arithmetic — and to make agent deployments that were marginal at old prices suddenly viable.
One caveat belongs in any honest account of this launch. OpenAI rates Terra as meeting its Critical cybersecurity threshold — its own classification for models capable enough in cybersecurity-relevant tasks to need extra safeguards. The release does not detail those safeguards, and the classification itself is OpenAI's, not an independent assessment. It is a reminder that cheaper, more capable agents cut both ways: they lower the cost of legitimate automation and of misuse alike.
Half the price, better at agent work — OpenAI just made the spreadsheet argument for agents much easier.
Why It Matters
First, price is becoming the decisive factor in enterprise AI, and this launch recognises it. Companies do not deploy agents because they are impressive; they deploy them because the maths works. Halving the per-token cost changes that maths for thousands of use cases that were previously too expensive to automate, and that is a bigger market expansion than any single capability jump.
Second, the distribution choice is smart. Launching on Bedrock the same day means Terra is available where enterprises already build, with their existing cloud contracts and compliance setups. A model that requires a new vendor relationship faces friction; a model inside AWS does not. That is why the Bedrock detail, easy to skip, is actually one of the most important sentences in the launch.
Third, the cybersecurity rating is a useful counterweight to the celebration. Cheaper agents mean more agents, and more agents mean more autonomous software touching real systems — which is exactly why identity, governance, and control are the other big enterprise AI stories this week. Terra's launch and the governance launches are two halves of the same story: capability is getting cheaper, so control must get better.
Analyst's Take
The verdict is bullish, straightforwardly. A flagship-class lab cutting agent prices in half while claiming better agent performance is good news for every company that wants to deploy agents at scale, and it pressures competitors to follow. The Bedrock availability on day one matters too: it means Terra reaches enterprises through Amazon's cloud, not just through OpenAI's own channels, widening its distribution from the start.
The watch items are verification and safety. OpenAI's benchmark and price claims are its own, and independent testing will show whether Terra really delivers GPT-6.1-class agent work at half the cost — or whether the savings come with capability trade-offs that matter in production. And the Critical cybersecurity rating is a flag, not a footnote: as agents get cheaper and more capable, the governance story told elsewhere in today's briefing — Oracle's governed runtime — becomes more urgent, not less. Bullish on the product, watchful on the consequences.
By The Numbers
FAQ
What is a token, and what does the price mean?
A token is a small chunk of text — roughly a word or part of a word — that AI models read and write. API prices are quoted per million tokens: GPT-6.1 Terra costs $1 for every million tokens you send in (input) and $5 for every million tokens it sends back (output). At those rates, processing a long document costs cents, not dollars.
Half the price of what?
Per OpenAI, half the price of GPT-5.6 Terra and GPT-6.1 Sol — the sibling models Terra is positioned against. The comparison is OpenAI's own; independent benchmarks will test whether Terra is truly as capable at half the cost.
What is Terra for?
OpenAI pitches it for computer-use workday tasks: forms, invoices, proposals, technical drawings, and video editing — the unglamorous office work that agents are increasingly asked to do. It is available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and to developers as gpt-6.1-terra.
How big is its context window?
1.05 million tokens — the amount of text and data the model can hold in mind at once. That is large enough to work across long documents and projects without forgetting the beginning. OpenAI says Terra is 15 percentage points better than GPT-5.6 Terra on Agents' Last Exam, a benchmark for agent tasks, at about a third of the per-task cost on that benchmark (not a third off the token price, which is halved).
Who can use it, besides OpenAI's own customers?
Amazon Web Services made it available on Bedrock the same day, so companies building on AWS can use Terra without going through OpenAI directly. OpenAI is private, so Microsoft, Nvidia, and Amazon are the public read-throughs: Microsoft as OpenAI's major partner, Nvidia as the chip supplier, Amazon as the distributor.
Is there a safety caveat?
Yes, and it is worth noting: OpenAI rates Terra as meeting its Critical cybersecurity threshold. That is OpenAI's own classification, and it means the model is considered capable enough in cybersecurity-relevant tasks to warrant extra safeguards. The release does not detail those safeguards.
👁
Source: PYMNTS