GPT-6.1 Terra: pricing, limits and the long-prompt surcharges

1. What GPT-6.1 Terra is#

GPT-6.1 Terra is the model OpenAI describes as fast and efficient, for visual reasoning, precise computer use and natural writing. Its release notes place it on October 26, 2026; the model page does not state the date. It is called in the API as gpt-6.1-terra.

The table gathers the data from its official page. Two details for developers: the maximum input (922,000 tokens) is smaller than the total window (1,050,000), because the output shares the space, and computer use works as a tool (computer_use) only in the Responses API.

Documentation: OpenAI: gpt-6.1-terra model page ↗

2. What it costs#

The Standard rate is $1 per million input tokens and $5 output. Input served from cache costs $0.05. Writing to the cache costs $1.25 with 30-minute retention or $2.50 with 12 hours. Batch and Flex are half price and fast mode double.

The rule that surprises most: if a request exceeds 272,000 input tokens, the input and cache rates double and the output rate is multiplied by 1.5 on the whole request, not only on the tokens above the limit. And regional processing, where available, adds 10%.

Documentation: OpenAI: API pricing ↗ · OpenAI: model release notes ↗

3. Calculate the cost of a request#

The function applies those rules. A request with 100,000 input tokens and 2,000 output costs $0.11; with 300,000 input it rises to $0.615 even though the output is the same, because the surcharge applies to everything; with 80% of the input served from cache it drops to $0.034; and in Batch, to $0.055. The documentation does not say how Batch, Flex or fast mode combine with a long prompt, so the function refuses to calculate it instead of making it up.

LONG_PROMPT_LIMIT = 272_000      # input tokens

def cost(input_tokens, output_tokens, cached=0, mode="standard"):
    """USD of a request to gpt-6.1-terra, at the published rates per million tokens."""
    long_prompt = input_tokens > LONG_PROMPT_LIMIT
    if long_prompt and mode != "standard":
        raise ValueError("the documentation does not say how Batch, Flex or fast mode combine with a long prompt")
    rate_in, rate_cache, rate_out = 1.00, 0.05, 5.00
    if long_prompt:                      # the whole request: 2x input and cache, 1.5x output
        rate_in, rate_cache, rate_out = rate_in * 2, rate_cache * 2, rate_out * 1.5
    factor = {"standard": 1.0, "batch": 0.5, "flex": 0.5, "fast": 2.0}[mode]
    usd = ((input_tokens - cached) * rate_in + cached * rate_cache + output_tokens * rate_out) / 1_000_000
    return round(usd * factor, 6)

print(cost(100_000, 2_000))                       # a normal request
print(cost(300_000, 2_000))                       # over 272K: surcharge on the whole request
print(cost(100_000, 2_000, cached=80_000))      # 80% of the input from cache
print(cost(100_000, 2_000, mode="batch"))         # Batch: half price

4. What a month costs#

With an example load of 150 million input tokens and 30 million output tokens a month, with no caching and short prompts, GPT-6.1 Terra costs $300, against $660 for GPT-5.6 Terra, $30 for GPT-6 Luna and $600 for Gemini 4 Argon at its introductory rate (which doubles on December 1). In pesos, with the TRM of October 28, 2026: COP 944,406 for GPT-6.1 Terra.

from decimal import Decimal as D

MODELS = {                                    # USD per million tokens (input, output)
    "GPT-6 Luna": (D("0.1"), D("0.5")),
    "GPT-6.1 Terra": (D("1"), D("5")),
    "GPT-5.6 Terra": (D("2"), D("12")),
    "Gemini 4 Argon (until Nov 30)": (D("2"), D("10")),
}
TRM = D("3148.02")                             # TRM of Oct 28, 2026
INPUT_M, OUTPUT_M = D(150), D(30)            # one month: 150 M input tokens and 30 M output tokens

for name, (t_in, t_out) in MODELS.items():
    usd = INPUT_M * t_in + OUTPUT_M * t_out
    print(f"{name}: ${usd:,.2f} a month | COP {usd * TRM:,.0f}")

5. Usage limits#

The per-tier limits from the model page, which rise automatically with spend and usage:

6. When it fits and when it does not#

• It fits when you need visual reasoning or computer use at a lower cost than OpenAI's largest models.
• It fits if your prompts are almost always below 272,000 tokens: there the rate is the list price.
• Beware of very long prompts: the surcharge applies to the whole request, and a large file can double your bill.
• For workloads that tolerate waiting, Batch and Flex cut the cost in half.
• Measure quality with your own tasks: prices do not say which answers better, and this guide does not compare quality.

7. Keep your numbers current#

The figures are those of OpenAI's official pages on October 28, 2026. Rates and surcharges change: confirm on the pricing page before budgeting.

• Check the pricing page every month.
• Measure what fraction of your requests exceeds 272,000 tokens.
• Recalculate in pesos with the TRM of the day you budget.