GPT-6.1 Terra API Guide: Visual Reasoning, Pricing and Tool Calls
GPT-6.1 Terra is an OpenAI model for visual reasoning, computer use and natural-language writing. It accepts text and images and produces text. Its official model ID is gpt-6.1-terra. The Guishu Intelligence global catalog also offers it; check the current pricing and specifications before connecting. OpenAI model documentation
Specifications and prices at a glance
| Item | Details |
|---|---|
| Model ID | gpt-6.1-terra |
| Input and output | Text and image input; text output |
| Official context window | 1,050,000 tokens; maximum output 128,000 tokens |
| Guishu catalog context | 1,000,000 tokens; current catalog and service configuration determine limits |
| Tool calls | OpenAI requires the Responses API |
| Official standard text pricing | Per million tokens: $1 input, $0.05 cached input, $5 output |
| Checked | 2026-10-28; prices and specifications may change |
Official direct pricing differs from Guishu billing. On 2026-10-27 our public catalog listed ¥0.8 input, ¥4 output and ¥0.08 cached input per million tokens. This is a dated snapshot; check the current price page and your actual bill before use. OpenAI adjusts long-context pricing for requests over 272K input tokens; tools and service modes may incur additional charges. See official model pricing. Do not apply OpenAI's long-context rules directly to other providers.
What tasks suit the model?
OpenAI lists matching product images with catalogs, interpreting technical drawings, operating application interfaces, and writing or editing text. Image input allows the model to answer using screenshots, charts or photos. Apply business rules and human review to results: check drawing units, product models and key fields, and do not execute high-risk actions solely on a model description.
Recognizing a screen and suggesting an action does not grant operating permission. The application must provide an execution environment, limit available pages and actions, and require confirmation for critical steps such as deletion, payment or sending. Results depend on screenshot quality, interface changes and tool implementation.
Connecting to the Responses API
This direct OpenAI text example requires an OpenAI API key and incurs charges. For third-party gateways, consult their documentation and catalog for the base URL, key and supported features; do not give an OpenAI key to a third party.
import os
from openai import OpenAI
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
response = client.responses.create(
model="gpt-6.1-terra",
input="List steps to check whether a product image matches its catalog entry.",
reasoning={"effort": "medium"},
)
print(response.output_text)
The official documentation lists reasoning.effort values none, low, medium, high, xhigh and max, with medium as the default. Compare accuracy, response time and usage on a fixed sample before selecting a level; the highest setting is not necessarily best.
Image input versus tool calls
Image input gives the model an image to interpret. A tool call is a structured request proposed by the model, executed by the application or a platform supporting the tool, and returned to the model. Tool calls require the Responses API. Supported tools, parameters and additional fees depend on the platform and configuration. OpenAI Responses API documentation
For screenshot analysis, first request a description and verifiable fields. For page operation, separately design executable actions, permissions, failure handling and confirmation. One successful text answer does not validate a complete automation workflow.
Estimating cost and testing results
- Prepare real tasks with easy, complex and error-prone examples, removing sensitive information.
- Keep prompts, image dimensions and inputs fixed; record accuracy, manual corrections, response time and actual input, cached and output tokens.
- Calculate input, cached input and output costs using the provider's current prices; check its rules for image, tool, long-context and other charges.
- Pilot in low-risk situations, then restrict permissions or add human review according to error types. Avoid logging keys or unnecessary private data.
For example, using our 2026-10-27 text prices above, one million ordinary input tokens plus 200,000 output tokens costs ¥0.8 + ¥0.8 = ¥1.6 in basic text fees. This excludes cache, images, tools and other fees; actual settlement depends on current rules and the bill.
FAQ
Can GPT-6.1 Terra generate images?
The model itself takes image input and produces text. OpenAI documents support for an image-generation tool in the Responses API; that is distinct from the model directly outputting images, and availability depends on service and tool configuration.
Why do OpenAI's 1,050,000 tokens and the catalog's 1,000,000 tokens differ?
The first is the window in OpenAI's model documentation; the second is the service specification in Guishu's catalog. Plan inputs against the current service limit, leaving room for output and conversation state.
Do tool calls guarantee operation of any website?
No. The application must supply the corresponding tools, execution environment and permissions. Verify support and safety boundaries for each operation.
See GPT-6.1 Terra pricing and integration or the official OpenAI model documentation for current specifications.