GPT-6.1 Terra vs GPT-6.1 Sol: price, specs and which to pick

GPT-6.1 Terra is OpenAI's cheaper, faster model at $1 per million input tokens and $5 per million output tokens, half of GPT-6.1 Sol's $2 and $10, with the same 1,050,000-token context window. OpenAI says to keep Sol for the most complex coding work and use Terra for writing, visual work, computer use and scoped code fixes. Last verified 2026-10-26 (America/Chicago).

GPT-6.1 Terra vs GPT-6.1 Sol at a glance

Both models are in OpenAI's GPT-6.1 line, take text and images, return text, and share a 1,050,000-token context window and 128,000-token output limit. The differences are price, release date, knowledge cutoff, which reasoning-effort levels exist, and where each is offered. Full facts: GPT-6.1 Terra and GPT-6.1 Sol.

Sources for the table: OpenAI's Terra model page, Sol model page and API changelog. Prices are standard processing in US dollars.

Cost per task, not cost per token

Terra's list price is exactly half of Sol's on every token type. That does not make every job half the price, because a model that needs more tokens, retries or a higher reasoning effort can erase the gap. The arithmetic below is illustrative: it assumes both models use the same number of tokens, which OpenAI does not claim.

Reported by one user, not verified by us: an X post on 2026-10-26 by EnvolDev gives one run of the same prompt as Opus 5.5 at $31.85 and 2h 45m, Sol at $3.62 and 43m, and Terra at $4.90 and 43m. The post states no reasoning-effort settings or task description beyond a prompt name, it is a single run, and it shows Sol cheaper than Terra on that run. Treat it as a reminder to measure your own task, not as a result.

Formula to use: cost per completed task = (input tokens x input price + output tokens x output price) per attempt, times the average number of attempts needed to pass your own check. Anthropic's cost and intelligence guide makes the same point about measuring completed tasks.

What OpenAI says about the difference

These are OpenAI's own claims from its launch post (read through the Internet Archive because openai.com blocks scripted fetches), its model-selection guide and its Codex models page. We have not reproduced any of them.

• Speed: the launch post says Terra completes tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol.
• Price: the launch post says Terra's standard API prices are 50% lower than GPT-5.6 Terra's and GPT-6.1 Sol's.
• Coding: the launch post gives Terra 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, and says "For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice." We did not find a Sol figure for the same benchmark in the post's text.
• Science agents: on Terminal-Bench-Science 0.1 at max reasoning, the post says Terra surpasses Sol at lower cost.
• Where to start: OpenAI's model-selection guide suggests Terra for concise writing from supplied material, computer-use tasks with a defined outcome and scoped code fixes, and Sol for complex projects where cost matters, such as a board presentation from financial results or a website from a product brief. It tells you to compare Terra with Sol on the same inputs and reasoning effort.

Details for Terra are in our GPT-6.1 Terra news brief.

Where each model is available

OpenAI's launch post also says that in ChatGPT Work and Codex the same usage allowance covers more tasks with Terra than with Sol or Astra. For a full tool-by-tool matrix see which coding agents can use GPT-6.1 Terra.

Which should you pick?

These are our editorial rules of thumb built on OpenAI's guidance, not test results.

• Start with Terra when the task is scoped, has a way to verify the result (tests, a saved record, a checklist), and you run it often enough that half the per-token price matters.
• Start with Sol when the task is long, ambiguous or multi-file, when a wrong attempt is expensive, or when OpenAI's own guidance says to keep Sol, such as the most complex coding work.
• Escalate rather than guess: run the same inputs on Terra first at the same reasoning effort, and move up to Sol only on the tasks that fail your check.
• Check the 272K-token line: above 272,000 input tokens both models bill the whole request at 2x input and 1.5x output, so the 2:1 price ratio holds but both cost more per request.
• If you also consider Anthropic's models, see GPT-6.1 Terra vs Claude Opus 5.5 and our best LLM for coding page.

What we have not tested

We have not run either model for this page. No speed, quality, token-efficiency or cost-per-task number here is ours. Benchmark names and scores are OpenAI's, from its own post, and OpenAI's charts are not copied because their values are not in the page text.

Latest news briefs on these models

• OpenAI GPT-6.1 Terra: a $1/$5 computer-use model, and what the launch post leaves out (Reported; first seen 2026-10-26 19:05 UTC)

FAQ

Sources

• OpenAI API: GPT-6.1 Terra model page (fetched 2026-10-26)
• OpenAI API: GPT-6.1 Sol model page (fetched 2026-10-26)
• OpenAI API changelog (fetched 2026-10-26)
• OpenAI: model selection guide (fetched 2026-10-26)
• OpenAI: Introducing GPT-6.1 Terra (Internet Archive capture; openai.com blocks scripted fetches) (fetched 2026-10-26)
• OpenAI: Codex and ChatGPT Work models (fetched 2026-10-26)
• GitHub changelog: GPT-6.1 Terra in GitHub Copilot (fetched 2026-10-26)
• GitHub Docs: supported AI models in Copilot (fetched 2026-10-26)
• Anthropic: optimizing for cost and intelligence (fetched 2026-10-26)
• EnvolDev on X: Opus 5.5 vs Sol vs Terra (one user's unverified run) (fetched 2026-10-26)

Related

• GPT-6.1 Terra: facts and sources
• GPT-6.1 Sol: facts and sources
• Which coding agents can use GPT-6.1 Terra
• Best LLM for coding
• Codex CLI

Page history

• Oct 26, 2026: First published. OpenAI model pages, API changelog, Codex models page and GitHub's Copilot changelog read on this date; nothing was run.