GPT-6.1 Terra pricing, API access and Sol benchmarks

OpenAI announced GPT-6.1 Terra on October 26, 2026, and the model is available through the API as gpt-6.1-terra. As of November 4, 2026, OpenAI's standard API price is $1 per million input tokens and $5 per million output tokens. That is half the uncached token price of GPT-6.1 Sol and 10 times the price of GPT-6 Luna.

The price list gives you a starting point. A lower price per token can still mean a higher cost per accepted task if Terra uses more tokens or needs more retries.

What is GPT-6.1 Terra for?

OpenAI positions Terra for work including visual reasoning, computer use and writing. It accepts text and images as input and returns text, according to the model reference. A screenshot can therefore be part of a request.

Terra's context window is 1,050,000 tokens, with a maximum output of 128,000 tokens. It has six reasoning effort settings: none, low, medium, high, xhigh and max. The default is medium. Reasoning effort is another setting to record when you compare costs and completion times.

Beyond the API, OpenAI says Terra is available in ChatGPT Work and Codex to Plus, Pro, Enterprise and Edu users.

What does GPT-6.1 Terra cost?

The table shows OpenAI's standard prices in US dollars per million tokens. Cached input is input served from a prompt cache, where a previously processed portion of a prompt can be reused.

Terra's uncached input and output rates are each 50% lower than Sol's; GPT-6 Luna's uncached rates are one tenth of Terra's. See our LLM API pricing comparison for rates from other providers.

The provider endpoint changes the price too. Requesty's OpenAI Terra endpoint lists upstream rates of $1 input and $5 output per million tokens. Its listing says pay as you go adds 5%, while bringing provider keys adds 0%.

The Azure West Europe Terra endpoint lists $1.20 input and $6 output before the pay-as-you-go fee. Requesty separately lists Terra Flex at $0.50 input and $2.50 output before the fee. Check the endpoint and service tier when comparing bills.

How does Terra compare with Sol and Luna?

The table below uses standard uncached OpenAI prices. The testing choices are starting points. You will need to measure the answers on your own tasks. The lower-priced comparison model is GPT-6 Luna, not GPT-6.1 Luna.

What do the benchmarks say?

OpenAI reports that Terra scored 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, compared with 21.5% for its predecessor, GPT-5.6 Terra. The predecessor's effort setting is unspecified in OpenAI's announcement.

For a Terra versus Sol comparison, Vals AI reports 49.49% for Terra and 55.05% for Sol on Terminal-Bench 4.0 using Mini-SWE-agent. Vals says it ran Terra through OpenAI at max. Sol's model profile lists max as its default, but says individual benchmarks can use different parameters, so its effort setting for this result remains unconfirmed.

Artificial Analysis reported Intelligence Index v5.0 scores of 51 for Terra and 55 for Sol, both at max, when checked on November 4, 2026. Test both models against your own acceptance criteria before choosing one.

Reasoning effort can change token use and completion time. Measure each effort setting you plan to use on the same tasks, with the same acceptance criteria.

How fast is Terra?

OpenAI says Terra completed tasks more than three times faster on average than GPT-6.1 Sol in its discussion of Agents' Last Exam, which tests professional computer-use tasks. That is a task completion result from OpenAI's evaluation, not API latency.

For API latency, Requesty's GPT-6.1 Terra latency ranking shows median and p95 latency per provider over the last 24 hours, from live production traffic.

For the earlier models, see our guide to GPT-6 Sol and Luna pricing and API access.

How do you access Terra through the API?

OpenAI identifies the direct API model as gpt-6.1-terra. Requesty's OpenAI deployment uses openai/gpt-6.1-terra, while the model-wide gpt-6.1-terra identifier routes across providers.

1. 1
   Choose an API

   For direct OpenAI access, use gpt-6.1-terra. For Requesty, use https://router.requesty.ai/v1 as the base URL and openai/gpt-6.1-terra to select the OpenAI deployment.

2. 2
   Follow the integration instructions

   OpenAI's Terra model reference lists the supported endpoints. Requesty's quickstart explains its OpenAI-compatible integration, and its Terra endpoint listing shows the deployment identifier.

3. 3
   Test the request on your workload

   Record the endpoint, reasoning effort and token counts alongside the result. Use the same acceptance criteria when you test another model.

OpenAI says Terra tool calling requires the Responses API. Requesty supports it through its Responses endpoint with openai/gpt-6.1-terra:

import os
from openai import OpenAI
 
client = OpenAI(
    base_url="https://router.requesty.ai/v1",
    api_key=os.environ["REQUESTY_API_KEY"],
)
 
response = client.responses.create(
    model="openai/gpt-6.1-terra",
    input="What is the weather in Paris?",
    tools=[{
        "type": "function",
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    }],
)
 
print(response.output)

What would Terra cost on a workload?

Take a hypothetical application that sends 10 million uncached input tokens and receives 2 million output tokens per month. Our calculation assumes the same token volume for each model, standard rates, and requests at or below Terra's 272,000-input-token threshold.

It excludes cache reads and writes, tool charges, regional premiums, retries and other charges. Direct prices come from OpenAI; the Requesty calculation uses its OpenAI Terra endpoint and assumes sufficient credit top-up.

Compare accepted results as well as token volume. If Luna needs more attempts to produce an accepted result, its cost per accepted task rises. At these prices and the same token volume per attempt, 10 Luna attempts would cost the same as one Terra attempt. Whether either model produces an acceptable result in those attempts is a separate question.

The same check applies to Terra and Sol. If Terra uses twice the assumed input and output volume to complete the same work, its direct cost rises to $40, equal to Sol's original estimate. How often either model needs a retry remains a question for your workload.

Endpoint choice changes the estimate as well. Using the listed Azure West Europe rates, the same 10 million input and 2 million output tokens cost $24 upstream, or $25.20 with a 5% fee, before any other applicable charges.

For more on that last measure, read cost per task, not cost per token.

Should you choose Terra?

Use Requesty's analytics headers to tag comparable workloads and examine spend and latency. Measure retries and accepted results in your application too. If you configure a fallback policy, test the fallback model on the same job before relying on its output.

For help designing the measurement, see our gateway latency and uptime comparison.