OpenAI Releases GPT-6.1 Terra, Standard API Price Cut 50% Compared with GPT-5.6 Terra

OpenAI has launched another new model in the GPT-6 series, focusing on completing everyday office and computer-operation tasks at lower cost.

On October 26, US Eastern Time, OpenAI released GPT-6.1 Terra, highlighting visual understanding, cross-application operation and natural-language writing capabilities. While upgrading model performance, it cut the standard API price by 50% compared with the previous generation, GPT-5.6 Terra. Evaluations by the independent testing organization Artificial Analysis also show that the new model improves in intelligence level, task cost and hallucination performance on specific benchmarks.

Designed for the entire workday, with stronger visual understanding and computer operation

OpenAI positions GPT-6.1 Terra as a fast, efficient model aimed at helping users complete computer-operation tasks across the entire workday at lower cost. The model can not only understand screen content, but also execute operations across applications and write in clearer, more natural language.

From drafting business proposals and interpreting technical drawings to filling out forms and editing videos, OpenAI hopes to make Terra a general-purpose tool suited to high-frequency office tasks. For enterprises, such capabilities also mean the model can take on more repetitive work, such as reading incoming invoices and then entering the relevant information into accounting software.

This upgrade does not merely pursue improvement in a single capability, but seeks to integrate visual understanding, computer operation and text generation into the same workflow. OpenAI says GPT-6.1 Terra has made clear progress over GPT-5.6 Terra on computer-operation and communication tasks, and also emphasizes that the new model allows the existing usage quotas of ChatGPT Work and Codex to complete more tasks.

Computer-operation benchmark up 15 percentage points, more than three times faster than Astra

On the Agents' Last Exam benchmark, which measures how agents use software to complete professional tasks, OpenAI says GPT-6.1 Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra, while task cost is about one-third of the latter's. Its performance can even match the flagship model GPT-6 Astra, yet the price per token is only one-tenth of Astra's.

At the highest reasoning effort, OpenAI says GPT-6.1 Terra's token efficiency is more than three times that of Claude Opus 5.5, and its price per token is only one-quarter of the latter's. Combining token-consumption efficiency with the unit-price advantage, Terra's cost to complete a task is about one-ninth of Claude Opus 5.5's. On speed, OpenAI says Terra completes tasks on average more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

In visual understanding, on the MMMU-Pro benchmark, GPT-6.1 Terra outperforms GPT-5.6 Terra under all reasoning settings and approaches the results of GPT-6 Astra and GPT-6.1 Sol at lower cost. On the OSWorld 2.1 benchmark, which tests agents using desktop applications to complete multi-step tasks, Terra at medium reasoning effort also surpasses GPT-5.6 Terra's previous best result.

These results reflect OpenAI's positioning of Terra: it does not necessarily pursue flagship-level performance across all capability dimensions, but rather seeks to complete a large amount of practical work at lower cost and in less time.

Independent evaluation: higher intelligence level, sharply lower task-completion cost

An independent evaluation released by Artificial Analysis further compared the performance and cost of GPT-6.1 Terra with the previous-generation model. The organization says Terra's top reasoning-effort score on the Artificial Analysis Intelligence Index is 6 points higher than GPT-5.6 Terra's, while the cost per individual task fell from $1.40 to $0.40, less than one-third of the original.

At lower and medium reasoning effort, Terra's index-score improvement over the previous generation is even larger, reaching 7 to 13 points. Artificial Analysis believes these two reasoning tiers sit on the efficiency frontier in the trade-off between intelligence level and cost, meaning users can obtain better model performance at relatively lower expense.

On the Coding Agent Index v2.0, aimed at coding agents, GPT-6.1 Terra scores 48 points, below GPT-6.1 Sol's 54 points but above GPT-6 Luna's 31 points. The organization notes that some of Terra's reasoning settings on this index are also on the cost-efficiency frontier, showing that it is not only suited to everyday office work but also has strong coding-agent capabilities.

However, the model's reasoning effort and task complexity still matter. Artificial Analysis's evaluation shows that Terra's Intelligence Index score at the highest reasoning effort is still lower than GPT-6.1 Sol's; OpenAI also explicitly recommends that for in-depth research or complex coding tasks, users should still prioritize GPT-6.1 Sol or GPT-6 Astra.

Hallucination rate drops sharply, coding task success rate more than doubles

Reliability is another key focus of this upgrade. In Artificial Analysis's private knowledge and hallucination benchmark AA-Omniscience, when the model does not know the answer, GPT-6.1 Terra gives a guess 38% of the time, compared with 88% for GPT-5.6 Terra, a drop of 50 percentage points. At the same time, Terra's accuracy on this benchmark also improved by 2 points.

It should be noted that this result measures the model's tendency to guess answers under uncertainty in a specific benchmark, and does not mean that the model's overall error rate in all practical applications has dropped by 50 percentage points.

OpenAI's own internal tests also show a similar trend of improvement. In prompt tests that easily induce errors, at low reasoning effort, GPT-6.1 Terra's error rate fell from 53.6% for GPT-5.6 Terra to 16.7%, a drop of 36.9 percentage points; at the highest reasoning effort, the error rate dropped by 18 percentage points to 4.3%, close to the level of GPT-6 Astra and GPT-6.1 Sol.

Coding capability has also improved markedly. In OpenAI's published Terminal-Bench 4.0 benchmark, GPT-6.1 Terra's task success rate at the highest reasoning effort reached 50.3%, exceeding the 21.5% previously published for GPT-5.6 Terra, more than doubling it. Artificial Analysis's independent evaluation likewise shows that Terra's score on this benchmark improved by about 20 percentage points, and at the highest reasoning effort it approaches GPT-6.1 Sol.

This means Terra's upgrade is not just about making the model faster and cheaper, but also about reducing some errors and improving its ability to execute complex tasks. However, for the most complex software engineering tasks, OpenAI still recommends using Sol or Astra.

API price halved, targeting high-frequency enterprise automation tasks

Price is another core highlight of the GPT-6.1 Terra release. OpenAI's published standard API pricing is: $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. Compared with GPT-5.6 Terra, the standard API price is cut by 50% overall; compared with GPT-6.1 Sol, Terra's standard API price is also 50% lower.

The price comparison provided by Artificial Analysis shows that GPT-5.6 Terra's input and output token prices were $2 and $12 per million respectively, while the new model drops to $1 and $5. The cached input price falls from $0.20 per million to $0.05, a reduction of 75%. The organization also reports that the new model's cache write price is $1.25 per million.

In ChatGPT Work and Codex, OpenAI says the same usage allowance can support GPT-6.1 Terra in completing more tasks than GPT-5.6 Terra, and more tasks than GPT-6.1 Sol and GPT-6 Astra. For enterprise users who need to repeatedly read files, operate office software, or process large amounts of structured information, this cost difference may directly affect whether agents can be deployed at scale.

GPT-6.1 Terra is available starting October 26 to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and developers can also call it via API under the model name gpt-6.1-terra.

From the perspective of product division of labor, OpenAI is further refining the capability and cost tiers of the GPT-6 series: Terra focuses on high-frequency, repetitive work that requires computer operation, while Sol and Astra continue to target in-depth research and complex coding. As models improve their execution efficiency on specific tasks, the focus of competition is also shifting from simply comparing the upper limit of model capability to measuring the time, tokens, and total cost required to complete real work. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this is at the user's own risk.