What to know first
Put Terra on the shortlist when a workflow needs to understand images or screens, complete bounded computer actions, make scoped code fixes, or process high-volume text. It is not the default for every complex project: compare GPT-6.1 Sol or GPT-6 Astra on the same inputs when you need deep research, long-horizon planning, or demanding engineering work.
What is GPT-6.1 Terra? Start with the release facts
OpenAI describes GPT-6.1 Terra as a fast, efficient model focused on detailed visual reasoning, precise computer use, and clearer writing. Its official examples include matching product photos to catalog entries, interpreting technical drawings, acting on application screens, and drafting or revising text.
For model selection, Terra is aimed at workflows with clear inputs and acceptance criteria. OpenAI's model-selection guidance suggests it for concise writing from supplied material, computer-use tasks with a defined outcome, and scoped code fixes. For complex projects, compare it with GPT-6.1 Sol or Astra using the same inputs and measure quality, speed, and rework.
| Search question | Official answer |
|---|---|
| When did GPT-6.1 Terra launch? | OpenAI's announcement is dated October 2026 and says it is available starting that day. |
| What is the API model ID? | `gpt-6.1-terra`. |
| What is it built for? | Visual reasoning, computer use, natural writing, and scoped software-engineering tasks. |
| Is it the strongest model? | No. OpenAI positions Astra as its flagship, Sol for complex work with a cost tradeoff, and Terra for visual reasoning and computer use. |
Sources: OpenAI: Introducing GPT-6.1 TerraOpenAI API: Model selection
Availability, model ID, and context specifications
The release announcement says GPT-6.1 Terra is available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and that developers can access it through the OpenAI API. The model, tools, limits, and plan access shown in an account can still vary, so check the relevant product and account pages.
The developer model page lists a 1,050,000-token context window, 128,000 maximum output tokens, and a June 1, 2026 knowledge cutoff. Terra supports `none`, `low`, `medium`, `high`, `xhigh`, and `max` reasoning effort, with `medium` as the default.
| Item | Details |
|---|---|
| ChatGPT Work / Codex | Plus, Pro, Enterprise, and Edu, subject to the account's actual availability |
| API model ID | `gpt-6.1-terra` |
| Reasoning effort | none, low, medium (default), high, xhigh, max |
| Context window | 1,050,000 tokens |
| Maximum output | 128,000 tokens |
| Input and output modalities | Text and image input with text output; the model page lists audio and video as unsupported |
| Tools and interface | Tool calling requires the Responses API; the model page lists web search, file search, code interpreter, computer use, and other tools |
| Knowledge cutoff | June 1, 2026 |
Sources: OpenAI: Introducing GPT-6.1 TerraOpenAI API: GPT-6.1 Terra Model
How much does the GPT-6.1 Terra API cost? A reproducible example
The developer model page lists standard API rates of $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. These rates cover text tokens; tool calls, regional processing, fast mode, long-context adjustments, and Batch or Flex pricing need to be checked separately on the official pricing documentation.
Suppose one request contains 1,000,000 uncached input tokens, 500,000 cached input tokens, and 100,000 output tokens. The standard token charge is 1×1 + 0.5×0.05 + 0.1×5 = $1.525. This example excludes tool-call fees and other adjustments; the actual bill should come from usage records.
| Meter | Per million tokens |
|---|---|
| Input | $1.00 |
| Cached input | $0.05 |
| Output | $5.00 |
Token cost = input tokens ÷ 1,000,000 × 1 + cached input tokens ÷ 1,000,000 × 0.05 + output tokens ÷ 1,000,000 × 5
Example = 1,000,000 ÷ 1,000,000 × 1 + 500,000 ÷ 1,000,000 × 0.05 + 100,000 ÷ 1,000,000 × 5 = $1.525
What evidence supports the capability claims?
OpenAI's release announcement reports results from several internal or public benchmarks. They help explain the model's positioning, but they do not guarantee the same lift for every business workflow. Before launch, test your own inputs, tools, and acceptance criteria.
The announcement also reports that on its error-inducing-prompt evaluation, the low-effort error rate fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, while the maximum-effort error rate was 4.3%. These are OpenAI-reported evaluations and should not replace factuality checks and human review in your workflow.
| Evaluation or metric | What the announcement reports | How to read it |
|---|---|---|
| Agents' Last Exam, medium reasoning | 15 percentage points above GPT-5.6 Terra at about one-third the cost | A signal for professional computer-use efficiency, not a promise that every task costs two-thirds less. |
| Price comparison with GPT-6 Astra | Some capability is near Astra at about one-tenth the price per token | Relevant when cost matters in high-volume automation. |
| Task speed | The announcement says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol | Measure end-to-end latency under your network, tools, and concurrency. |
| Terminal-Bench 4.0, maximum reasoning | Terra 50.3% versus GPT-5.6 Terra's published 21.5% | A meaningful coding improvement, while complex engineering still merits a Sol/Astra comparison. |
Sources: OpenAI: Introducing GPT-6.1 Terra
How should you choose between Terra, GPT-6.1 Sol, Astra, and Luna?
Do not choose from the model name or one answer alone. Fix a representative task set, then record completion rate, review time, end-to-end latency, token cost, and rework after failures. OpenAI's model-selection guidance recommends keeping the lightest model and reasoning setting that meets your quality bar.
If your workflow is mainly screen understanding, image handling, bounded application actions, or small code changes, put Terra first in the trial. If it requires cross-file planning, complex technical deliverables, or deeper analysis, include Sol and Astra in the comparison.
| Model | Good starting point for | What to watch |
|---|---|---|
| GPT-6.1 Terra | Visual reasoning, computer use, concise writing, scoped code fixes | Define the supplied material, action boundary, and acceptance test; measure tools and latency. |
| GPT-6.1 Sol | Complex projects that still need a quality-cost balance | Compare it with Terra at the same inputs and reasoning effort. |
| GPT-6 Astra | The most demanding reasoning, research, and deliverables | Budget for the higher price and latency when quality is the priority. |
| GPT-6 Luna | Scoped lightweight tasks, triage, and frequent automation | Start here for cost-sensitive work; escalate when complexity requires it. |
Sources: OpenAI API: Model selectionOpenAI: Introducing GPT-6.1 Terra
What should you check before integrating the API?
- Set the request model to `gpt-6.1-terra`; do not use a ChatGPT subscription name as an API model ID.
- For tool calling, use the Responses API described in the developer documentation and define timeouts, permissions, and retry behavior.
- Record `reasoning.effort` as a tunable parameter. Start with the default medium setting, then compare low, high, or max on the same task set.
- Monitor image inputs, long contexts, and output limits so that visible text does not become a proxy for token usage.
- Run regression tasks before launch and record factual errors, whether computer actions were saved, code-test results, and human repair time.
- Build a budget from standard rates, then separately check caching, Batch/Flex, regional processing, fast mode, and tool-call rules.
Sources: OpenAI API: GPT-6.1 Terra ModelOpenAI API：PricingOpenAI API: Model selection
Frequently asked questions
What is the GPT-6.1 Terra API model ID?
The model ID is `gpt-6.1-terra`. Use that ID in the API request instead of a ChatGPT Plus, Pro, or other subscription name.
Sources: OpenAI API: GPT-6.1 Terra Model
How much does GPT-6.1 Terra cost per million tokens?
The developer model page lists $1 for input, $0.05 for cached input, and $5 for output per million tokens. Tool calls and other pricing modes need separate checks.
Is GPT-6.1 Terra good for complex coding projects?
It can handle scoped code fixes and software-engineering tasks, but compare it with GPT-6.1 Sol or GPT-6 Astra on the same task set for complex projects instead of deciding from one benchmark score.
Sources: OpenAI API: Model selectionOpenAI: Introducing GPT-6.1 Terra
Does GPT-6.1 Terra support images, audio, and video?
The official model page lists text and image input with text output, and marks audio and video as unsupported. Check the current model page for the latest modality list before integrating.
Sources: OpenAI API: GPT-6.1 Terra Model
Does a ChatGPT subscription include GPT-6.1 Terra API credits?
ChatGPT Work/Codex plan usage and developer billing on the OpenAI API are separate product paths. The announcement says eligible plan users can use Terra in ChatGPT Work and Codex; API charges are calculated from actual token and tool usage on the API platform.
Sources: OpenAI: Introducing GPT-6.1 TerraOpenAI API: GPT-6.1 Terra Model
Official sources and review notes
Sources checked on October 27, 2026. Rates and eligibility rules can change; confirm the latest official guidance and your own agreement.
- OpenAI: Introducing GPT-6.1 Terra
- OpenAI API: GPT-6.1 Terra Model
- OpenAI API：Pricing
- OpenAI API: Model selection
AISub prepared this guide from public OpenAI documentation, with an independent worked example and purchasing advice. AISub is an independent price index and is not affiliated with OpenAI.