GPT-6.1 Terra API Guide (2026): $1/$5 Pricing, 1.05M Context, and When to Switch from Sol or Sonnet

On October 26, 2026, OpenAI added GPT-6.1 Terra to its API. The model ID is gpt-6.1-terra, and it costs $1 per million input tokens and $5 per million output tokens. That's exactly half of GPT-6.1 Sol's $2/$10, and half of Claude Sonnet 5.5's $2/$10 too. The release is logged in OpenAI's API changelog, and independent numbers from Artificial Analysis landed the same day.

Short version: Terra is the model most teams should test first for high-volume work, and it isn't the model you hand your hardest coding agent. Details below, including one pricing rule that can quietly double a long-context bill.

Key Takeaways

What OpenAI Actually Shipped

GPT-6.1 Terra is a mid-tier reasoning model aimed at visual reasoning, computer use and writing. OpenAI's own examples: matching product photos to catalog entries, reading technical drawings, acting on application screens, and drafting or revising text. It accepts text and image input and returns text only. Audio, video and fine-tuning aren't supported. The knowledge cutoff is June 1, 2026.

Reasoning effort is a real dial here. The reasoning.effort parameter accepts none, low, medium (the default), high, xhigh and max. That matters for cost, because OpenAI bills reasoning tokens as output tokens at $5 per million.

The numbers that got attention, as reported on October 26 by The Decoder citing OpenAI and Artificial Analysis:

• Terminal-Bench 4.0: 50.3% at max effort, up from 21.5% for GPT-5.6 Terra.
• Intelligence Index: 48, versus 52 for GPT-6.1 Sol and 42 for GPT-5.6 Terra.
• Hallucination: on AA-Omniscience, Terra guesses 38% of the time when it lacks the answer, down from 88%.
• Cost per task at max effort: about $0.40, versus $1.40 for GPT-5.6 Terra.

Treat those as averages, not as your workload.

GPT-6.1 Terra API Pricing

All prices below are USD per 1 million tokens, taken from OpenAI's model page on October 28, 2026. Rows other than Standard apply the multipliers OpenAI lists on that page.

Two details to remember. Cached input costs 5% of the uncached rate, which is $0.05 per million. And the long-context rule applies to the whole request, not just the overflow: a 300K-token prompt is billed at $2 per million input tokens across all 300K tokens.

GPT-6.1 Terra vs GPT-6.1 Sol vs Claude Sonnet 5.5

The comparison most people want is Terra against the model it undercuts. Here's how the neighbors stack up on the attributes that decide a routing choice.

My read: Terra now owns the middle. Luna is too weak for agentic work, and Sol and Sonnet 5.5 cost double. Unless your workload is mostly hard coding, the burden of proof just moved to the expensive models.

What It Costs on a Real Workload

Take an invoice pipeline: 100,000 documents a month, about 3,000 input tokens and 400 output tokens each. That's 300 million input tokens and 40 million output tokens.

• GPT-6.1 Terra, Standard: $300 input plus $200 output, so $500 per month.
• GPT-6.1 Terra with a cached 2,000-token prefix: 200M cached tokens at $0.05 is $10, the other 100M input tokens are $100, and output is $200. That's $310 per month.
• GPT-6.1 Terra, Batch: $150 input plus $100 output, so $250 per month.
• GPT-6.1 Sol or Claude Sonnet 5.5, Standard: $600 input plus $400 output, so $1,000 per month.

That's a 4x gap between Sol and Terra on Batch, if you can live with batch turnaround. Plug your own token counts into the API cost calculator or measure real prompts with the token counter before you commit.

Calling gpt-6.1-terra: curl, Python, Node.js

Use the Responses API. Terra also works on Chat Completions, but OpenAI says tool calling requires Responses. A migration that only swaps the model string can pass a text-only smoke test and then fail the first time your app calls a tool.

curl "${OPENAI_BASE_URL:-https://api.openai.com/v1}/responses" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-terra",
    "reasoning": {"effort": "low"},
    "input": "Extract vendor, total and due date from this invoice as JSON: ..."
  }'

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["OPENAI_API_KEY"],
    base_url=os.environ.get("OPENAI_BASE_URL", "https://api.openai.com/v1"),
)

resp = client.responses.create(
    model="gpt-6.1-terra",
    reasoning={"effort": "low"},
    input=[{
        "role": "user",
        "content": [
            {"type": "input_text", "text": "Return vendor, total and due_date as JSON."},
            {"type": "input_image", "image_url": "https://example.com/invoice-1042.png"},
        ],
    }],
)

print(resp.output_text)
print(resp.usage)  # check reasoning tokens: they bill as output

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: process.env.OPENAI_BASE_URL ?? "https://api.openai.com/v1",
});

const resp = await client.responses.create({
  model: "gpt-6.1-terra",
  reasoning: { effort: "low" },
  input: "Classify this support ticket and draft a two-line reply: ...",
});

console.log(resp.output_text);

Start extraction and classification at low effort. Artificial Analysis reported that Terra's gains over GPT-5.6 Terra were largest at lower effort levels, 7 to 13 points, so you're not paying for max reasoning to get the improvement. Only raise it when your own evals say you need to.

Because the base URL is an environment variable, you can also A/B Terra against Sonnet 5.5 through an OpenAI-compatible gateway such as KissAPI without touching client code. Check your dashboard's model list for what's currently enabled.

When to Switch, and When Not To

Switch first: invoice and form extraction, screen-driven agents, support drafting, and anything where you're paying Sol prices for work that doesn't need Sol.

Hold off: multi-file refactors and long-running coding agents. OpenAI itself still points those at GPT-6.1 Sol and GPT-6 Astra, and the six-point Coding Agent Index gap is real. A sane setup is a router: default to Terra, escalate to Sol after repeated validation failures.

Two cautions. First, don't fill the context window just because you can. Crossing 272,000 input tokens reprices the whole request at 2x input and 1.5x output, so chunk or retrieve instead. Second, OpenAI's system card puts Terra's defense against indirect prompt injection at 98.8%, up from 96.2%. That's better, not solved. Keep your sandboxing.

Caching is the other lever. A 30-minute cache write costs $1.25 per million tokens and a 12-hour write costs $2.50, so long-lived prefixes need real reuse to pay off. The 12-hour prompt caching guide has the break-even math.

Frequently Asked Questions

How much does the GPT-6.1 Terra API cost?

GPT-6.1 Terra costs $1.00 per 1 million input tokens, $0.05 per 1 million cached input tokens, and $5.00 per 1 million output tokens at Standard rates. Batch and Flex processing are 50% lower, at $0.50 input and $2.50 output per 1 million tokens.

What is the context window of GPT-6.1 Terra?

GPT-6.1 Terra has a 1,050,000-token context window and a 128,000-token maximum output. Requests with more than 272,000 input tokens are billed at 2x the input and cache rates and 1.5x the output rate for the entire request.

Is GPT-6.1 Terra better than GPT-6.1 Sol for coding?

No. Artificial Analysis scored GPT-6.1 Terra 48 and GPT-6.1 Sol 54 on its Coding Agent Index v2.0, and OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the most demanding coding tasks. Terra costs half as much, so it fits routine and high-volume work better.

Sources: OpenAI API changelog entry dated October 26, 2026; the GPT-6.1 Terra model page on developers.openai.com (pricing, context and rate limits read on October 28, 2026); The Decoder, "GPT-6.1 Terra halves OpenAI's API prices and still gains ground on its larger models," October 26, 2026, citing OpenAI and Artificial Analysis.