OpenAI unveils GPT-6.1 Terra, cutting API prices in half

[Financial News] OpenAI unveiled GPT-6.1 Terra, an artificial intelligence (AI) model capable of performing a range of tasks, from drafting documents to interpreting technical drawings and editing videos. By improving performance over previous models while cutting usage costs in half, it aims to tap into businesses’ demand for AI-powered task automation.
OpenAI said on the 27th that GPT-6.1 Terra would be available through ChatGPT Work, OpenAI Codex and the OpenAI application programming interface (API).
GPT-6.1 Terra combines the ability to understand visual information and perform computer-based tasks with writing and coding skills. It can handle complex tasks such as drafting proposals, converting architectural drawings into three-dimensional (3D) models and recording timecodes for individual scenes in videos. It can also be used for repetitive tasks, such as reading invoice details and entering them into accounting software.

In particular, OpenAI has improved its performance relative to cost. Standard API prices are 50% lower than those for GPT-5.6 Terra and GPT-6.1 Sol. ChatGPT Work and Codex can also handle more tasks than before within the same usage limits.
It also showed improved results in performance evaluations. On Agent Last Exam, which assesses the ability to perform professional tasks, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at the medium reasoning level. The cost of completing tasks was about one-third as much. It also outperformed its predecessor across all reasoning configurations on MMMU-Pro, a visual reasoning benchmark.
It also reduced AI “hallucinations,” or the generation of information that does not match the facts. In an internal evaluation consisting of questions designed to induce errors, the error rate at the low reasoning level was 16.7%, 36.9 percentage points lower than the previous model’s 53.6%. At the maximum reasoning level, the error rate fell to 4.3%.
Coding performance also improved. On Terminal-Bench 4.0, which measures the ability to solve coding problems in a command-line environment, it achieved a solve rate of 50.3% at the maximum reasoning level. That was more than double the previous model’s 21.5%. OpenAI said, however, that GPT-6.1 Sol and GPT-6 Astra are better suited to highly demanding coding tasks.
In terms of safety, GPT-6.1 Terra reached the “Critical” threshold—the highest risk level—in a cybersecurity capability assessment under OpenAI’s Preparedness Framework. Accordingly, it was given the same safeguards as GPT-6 Astra. Its resistance to prompt-injection attacks, which use malicious instructions to manipulate an AI’s tasks, was also strengthened.
GPT-6.1 Terra will be available to users of ChatGPT Work and Codex, as well as ChatGPT Plus, ChatGPT Pro, ChatGPT Enterprise and ChatGPT Edu. API pricing per 1 million tokens is $1 for input, $0.05 for cached input and $5 for output.
[email protected] Choi Hye-rim Reporter