OpenAI Launches GPT-6.1 Terra, Its Half-Good AI Model at Half the Price
In brief
- OpenAI released GPT-6.1 Terra on October 26. Its standard API prices are 50% lower than those of GPT-5.6 Terra.
- It is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can also use it as gpt-6.1-terra.
- In OpenAI's internal test, the rate of factual errors at low reasoning effort dropped from 53.6% to 16.7%. At the highest effort setting, Terra completed 50.3% of Terminal-Bench 4.0 tasks, compared with 21.5% for GPT-5.6 Terra.
OpenAI released GPT-6.1 Terra on Monday. This is an improved version of the middle model in its lineup. OpenAI says it costs developers 50% less than GPT-5.6 Terra and can handle computer tasks faster and more accurately.
Terra is designed for everyday tasks, such as reading what is on a screen, clicking through apps, and writing replies. This ability is called computer use. It means the AI can look at your screen, click things, and type as a person would. The 50% price cut applies to standard API prices, which are the rates developers pay to connect the model to their own apps.
“It’s a big improvement over GPT-5.6 Terra, with standard API prices that are 50% lower,” OpenAI said in its announcement. The company added that “GPT-6.1 Terra is rolling out now and is available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.”
ChatGPT Work is OpenAI's agent mode inside ChatGPT. It can use different apps and browse websites and complete tasks for you. Codex is OpenAI's coding tool. Free users can use the lower-level GPT-5.6 Luna model.
OpenAI tested the models internally using prompts designed to trick them into making mistakes. With reasoning effort set to low, a setting that controls how long the model thinks, the rate of factual errors dropped from 53.6% for GPT-5.6 Terra to 16.7% for GPT-6.1 Terra. At the highest effort setting, the error rate fell by 18 percentage points, reaching 4.3%.
Terminal-Bench 4.0 tests AI coding agents by giving them tasks to complete in a computer's command-line terminal. It measures the percentage of tasks they finish successfully. At the highest effort setting, Terra completed 50.3% of tasks, compared with 21.5% for GPT-5.6 Terra. OpenAI also says Terra completes tasks more than three times faster on average than GPT-6 Astra and GPT-6.1 Sol. All these figures come from OpenAI's own tests, including the internal test of factual errors.
OpenAI first introduced Terra on June 26 as part of the GPT-5.6 family in a limited preview requested by the Donald Trump administration. On July 9, the company made it available to the public as a model that could compete with GPT-5.5 while costing about half as much.
In September, the company released GPT-6 Sol and Luna just minutes after Anthropic released Claude Opus 5.5. OpenAI did not release a new version of Terra at that time.
For detailed research or difficult coding tasks, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra.