GPT-6.1 Terra: $1/$5 Pricing, Benchmarks, and When to Pick It

OpenAI released GPT-6.1 Terra on October 27, 2026: a $1 input / $5 output per million token model with a 1,050,000-token context window, API id gpt-6.1-terra, aimed at computer use, visual reasoning and writing rather than the hardest coding. OpenAI says its Terminal-Bench 4.0 score at max effort is 50.3%, more than double GPT-5.6 Terra's published 21.5%, at half the standard price of GPT-6.1 Sol.

Official Sources#

What Shipped#

Per the model page, checked on October 27:

• Model id: gpt-6.1-terra. Text and image in, text out. Audio and video are not supported, and fine-tuning is not supported.
• Context: 1,050,000 tokens, with up to 128,000 output tokens and a June 1, 2026 knowledge cutoff.
• Reasoning: reasoning.effort accepts none, low, medium (default), high, xhigh and max.
• Tool calling: requires the Responses API. Computer use, MCP, skills, hosted shell, apply patch and tool search are listed as supported.
• Where: all Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, plus the API.
• Rate limits (Build tier): 5,000 requests per minute and 1,000,000 tokens per minute.

OpenAI's framing is "fast and capable computer use at scale": filling forms, reading invoices into accounting software, interpreting technical drawings. It is explicitly not pitched as the coding flagship. The announcement says that for the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice.

Benchmarks (all vendor-reported)#

Every number below comes from OpenAI's announcement. None is independently measured yet, and OpenAI's charts mix its own internal evals with public ones.

Two caveats. First, the Opus 5.5 comparison is OpenAI's own chart on OpenAI's chosen benchmark, so treat the 9x as directional. Second, the Terminal-Bench 4.0 headline is a jump from a weak base: our Opus 5.5 vs Sol vs Grok comparison tracked Anthropic reporting 66.4% for Opus 5.5 on the same benchmark, so 50.3% is a big step for Terra and still not the top of the table.

OpenAI also says Terra meets the Critical cybersecurity threshold under its Preparedness Framework and keeps the safeguards built for GPT-6 Astra. In practice that means more refusals and monitoring on security-adjacent prompts than a lower tier would have drawn.

Pricing and Context#

Terra prices come from the model page on the day of release. Sibling rows are the rates in our earlier same-day-verified posts; re-check before you commit budget.

Terra's rules from the model page: cache writes are $1.25 per MTok at 30-minute retention ($2.50 at 12 hours). Prompts over 272K input tokens bill at 2x input and cache rates and 1.5x output for the full request. Batch and Flex are 50% off, Fast mode is 2x, and regional processing adds 10%.

Worked cost per task#

Assume one agent task of 1,000,000 input tokens (900,000 of them cache reads, 100,000 uncached) and 60,000 output tokens, under the 272K per-request surcharge line. Token counts are our assumption and will differ per model in practice, since a stronger model may need fewer turns.

Terra is exactly half of Sol on every axis, so the question is only whether the quality gap is worth the 2x.

The angle: the middle child just got a real job#

On release day the loudest Hacker News reaction was not about a benchmark, it was about naming: Luna, Terra, Sol, Astra, now with version numbers and six effort levels, which one commenter counted as roughly 20 combinations. That confusion is the real cost for builders. A new tier release no longer answers "which model should I use?" - it adds one more cell to a routing table.

Who wins: teams running high-volume, vision-heavy or form-filling agents, where Terra's computer-use focus, $0.05 cache reads and 3x speed matter more than peak coding score. Who loses: Anthropic's mid-tier story and anyone selling "cheaper Sonnet-class" pricing, because Terra's $1/$5 sticker undercuts it while OpenAI explicitly benchmarks against Opus 5.5. The second-order effect is on OpenAI's own lineup. Terra at half of Sol, with Sol itself recently cut, shrinks the space where Sol is the obvious default, so expect routing guidance to shift toward "Terra first, escalate to Sol or Astra on failure."

What people are actually saying#

The HN thread (488 points, 310 comments at crawl time) split three ways:

• Positioning skeptics. One commenter called Terra a "weird middle child" between a much-improved small model and a much-cheaper large one, and said the Terminal-Bench 4.0 cost-per-task chart left them unsure why anyone would pick Terra over Sol for coding. Another read the release as a flex aimed at Anthropic's Sonnet 5.5.
• Production builders. A developer running an application almost entirely on the OpenAI API called Terra a strong release for cost-sensitive agentic production, while saying they still use Claude for coding. Another commenter argued software developers are not the intended audience at all; computer work, automation and technical writing are.
• The naming complaint and the OpenAI reply. An OpenAI employee in the thread explained that model size explains most of the performance ordering but that skills are multidimensional, and said they could not give a one-line answer for something like technical documentation. Several readers replied that this is exactly the problem, and the most common advice was to build a small eval on your own data.

The counter-case: if you only want cheap coding, commenters pointed to DeepSeek rather than Terra. That fits OpenAI's own text, which keeps Sol and Astra as the coding picks.

Run It Monday#

Codex. Terra is listed as available in Codex. The model flag pattern we verified earlier for Sol on codex-cli:

codex -m gpt-6.1-terra
codex exec -m gpt-6.1-terra "Fill the CSV from the invoice PDFs in ./inbox and list any you could not parse"

API. Tool calling needs the Responses API:

from openai import OpenAI

client = OpenAI()
response = client.responses.create(
    model="gpt-6.1-terra",
    reasoning={"effort": "medium"},
    input="Read the failing CI log, name the root cause, and patch it.",
)
print(response.output_text)

OpenCode. We could not confirm gpt-6.1-terra in the model registry OpenCode reads on release day, so there is no opencode run line to hand you. Check opencode models after updating; the pattern from our Sol guide applies once it lands.

A routing test that takes an afternoon. Take 30 real tasks from your own logs, run them at medium effort on Terra and on whatever you use today, and score pass rate and dollars per passing task. The routing strategies guide has the escalation pattern if Terra passes the easy bucket but fails the hard one.

Which One to Pick#

For the family overview, start with GPT-6 in 7 Minutes.

FAQ#

What is the API model id for GPT-6.1 Terra?#

It is gpt-6.1-terra, per OpenAI's model page. Use the Responses API for tool calling. Reasoning effort accepts none, low, medium (default), high, xhigh and max.

How much does GPT-6.1 Terra cost?#

Standard pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, which OpenAI says is 50% below GPT-6.1 Sol. Prompts over 272K input tokens bill at 2x input and 1.5x output for the whole request. Batch and Flex are 50% off.

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

No, according to OpenAI. The announcement says Sol and Astra remain the better choice for the most complex coding tasks. Terra's vendor-reported 50.3% on Terminal-Bench 4.0 at max effort is a large jump over GPT-5.6 Terra, but run your own tasks before routing coding work to it.

Can I use GPT-6.1 Terra in Codex?#

OpenAI says it is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex starting October 27. Select it with codex -m gpt-6.1-terra.

Sources#

Last updated: October 27, 2026

Continue Reading#

• GPT-6.1 Sol Release Guide: Near-Astra Agentic Work at $2/$10 and $0.10 Cache Reads - the tier Terra undercuts by half
• GPT-6 in 7 Minutes: Astra, Sol, and Luna Explained for Developers - the family map before the 6.1 releases
• Cheapest Subagent Model: Haiku 5.5 or GPT-6 Luna? Routing, Compaction and Cost per Task - the tier below Terra
• AI Model Routing Strategies for Cost-Effective Coding in 2026 - how to escalate between tiers