OpenAI announced GPT-6.1 Terra on October 26, 2026, positioning it for visual reasoning and computer use. The company said rollout was underway for Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and listed the API model as gpt-6.1-terra.
OpenAI positions Terra for visual reasoning and computer use
OpenAI describes Terra as a model that can understand what appears on a screen and take actions across applications. It also says the model writes naturally. The launch announcement framed the release as a way to do more work within the same usage allowance and said standard API prices were 50% lower than GPT-5.6 Terra.
What the API supports, and where it draws the line
GPT-6.1 Terra accepts text and image input and returns text. OpenAI lists a 1,050,000-token context window and a 128,000-token maximum output. A token is a unit of text processed or generated by the model; the number of tokens in a prompt depends on its content.
The API supports computer-use tools through the Responses API, and tool calling requires that API. OpenAI also lists tools such as code interpreter, hosted shell, and apply patch. The model’s reasoning setting can be none, low, medium (the default), high, xhigh, or max.
The documented input boundary matters for workflow fit: Terra accepts images, but not audio or video input. OpenAI also lists fine-tuning as unsupported. Its launch guidance points users to GPT-6.1 Sol or GPT-6 Astra for in-depth research and complex coding, even though Terra’s API documentation includes coding-related tools.
GPT-6.1 Terra API rates and billing conditions
OpenAI lists these standard API rates in U.S. dollars per million tokens:
| Token category | Standard price per 1 million tokens (USD) | Condition |
| Input | $1.00 | Standard rate |
| Cached input | $0.05 | Standard cached-input rate |
| Cache write | $1.25 | 30-minute retention |
| Cache write | $2.50 | 12-hour retention |
| Output | $5.00 | Standard rate |
A simple estimate makes the input-output split clear: 100,000 uncached input tokens cost $0.10, and 20,000 output tokens cost another $0.10, for a total of $0.20 at standard rates before tool-call fees or other pricing adjustments.
Long prompts can change the calculation. When a request exceeds 272,000 input tokens, OpenAI applies twice the input and cache rates and 1.5 times the output rate to the entire request. Batch and Flex rates are 50% below Standard, Fast mode costs twice the Standard rate, and regional processing carries a 10% premium where available. OpenAI’s GPT-6.1 Terra API documentation lists these rates and conditions.
Fast output does not guarantee a fast first response
OpenAI said Terra completed tasks more than three times faster on average than GPT-6 Astra and GPT-6.1 Sol. It also reported a 50.3% score for Terra on Terminal-Bench 4.0 at maximum reasoning effort, compared with 21.5% for GPT-5.6 Terra. These are OpenAI’s stated comparisons; the benchmark score and task-speed claim describe different measures.
A review reporting Artificial Analysis measurements puts another distinction in focus. At maximum reasoning effort, it reports about 90.9 output tokens per second for Terra, alongside 131.81 seconds to the first token. At low effort, the same account gives 91.0 tokens per second and 1.29 seconds to the first token. Output rate measures how quickly tokens are generated once a response begins; time to first token measures the wait before that starts.
AI Coding Daily reported different timings in its own custom coding test: about 26 seconds per prompt at medium or high reasoning effort, and 7 to 8 minutes at maximum effort. It put maximum-effort API costs at about $0.24 to $0.27 per prompt. Those figures belong to that channel’s test setup, not to a general price or response-time guarantee.
Individual Reddit users described mixed experiences with longer tasks. Some reported progress interruptions or a task stopping during PDF-based work; another praised Terra on routine office and music-library tasks. These were individual launch-period accounts.
