OpenAI's GPT-6.1 Terra: Half the Price, Built to Actually Use Your Computer

Nothing here is sponsored — no affiliate links on this page. Every benchmark figure below is OpenAI's own measurement from its launch announcement, not an independently verified result. We have not run the model ourselves — this is researched analysis, not hands-on testing.

OpenAI released GPT-6.1 Terra on October 26, 2026, and the headline is refreshingly boring in the best way: it is cheaper, faster, and built for the unglamorous work that actually fills a developer's day. Not a flagship flex. A workhorse price cut for the model that clicks buttons, reads forms, and turns drawings into documents. If you run agents that touch real screens and real documents at volume, this is the model change that moves your bill.

What GPT-6.1 Terra actually is

Terra replaces GPT-5.6 Terra in OpenAI's lineup. It sits in the middle: above GPT-6 Luna, the cheapest model, and below GPT-6.1 Sol and the flagship GPT-6 Astra.

It is a reasoning model built for high-volume visual and computer-use work. It takes text and image input, has a 1.05-million-token context window with up to 128,000 output tokens, and it can operate a computer: look at a screen, click, type, and move between applications the way a person would.

It is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra. Amazon Web Services made it available on Bedrock the same day. It is not in regular Chat, and no free tier is listed.

Through the Responses API it supports tools including computer use, web search, file search, code interpreter, hosted shell and MCP. OpenAI's own example workloads tell you exactly who it is for: drafting proposals, interpreting technical drawings, filling out forms, entering invoices into accounting software, matching product photos to catalogue entries, and editing video by scene.

The price is the story

Standard API pricing per million tokens:

• GPT-6.1 Terra: $1 input, $0.05 cached input, $5 output
• GPT-5.6 Terra (the model it replaces): $2 input, $12 output
• GPT-6.1 Sol: $2 input, $10 output
• GPT-6 Astra (flagship): $10 input, $50 output

So Terra is half the price of both its predecessor and Sol, and roughly one-tenth the token price of Astra. Independent developer write-ups put it the same way: almost half the price of GPT-5.6 Terra and GPT-6.1 Sol, and around 10x the price of GPT-6 Luna.

Two pricing details matter if you run long or batched workloads. First, the long-prompt cliff: prompts longer than 272,000 input tokens are billed at 2x the input and cache rates and 1.5x output — for the whole request, not just the part over the line. Second, Batch and Flex processing are 50% cheaper than Standard, while Fast mode costs 2x Standard. For overnight, non-urgent agent batches, the Batch/Flex route halves the bill again.

What OpenAI claims it can do

Again: these are OpenAI's claims, not independent results. But the direction is consistent, and it is the reason this launch matters.

On Agents' Last Exam — professional tasks like turning architectural drawings into a 3D model — OpenAI says Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and performs on par with GPT-6 Astra at only one-tenth its price per token. It also says Terra finishes tasks more than three times faster than Astra or Sol. At maximum effort, OpenAI says Terra is over three times as token-efficient as Anthropic's Claude Opus 5.5 at a quarter of its price per token — roughly a 9x lower cost per task.

On Terminal-Bench 4.0, Terra solved 50.3% of tasks at maximum effort, up from 21.5% for its predecessor. On an internal set of prompts designed to trigger errors, the error rate at low reasoning effort fell from 53.6% to 16.7%. On OSWorld 2.1, which tests desktop-application use, Terra beat its predecessor's best score at medium reasoning effort while costing less per task.

The honest read: cheaper and more capable at once, on the vendor's own tests. Independent testing will tell us how much of that survives contact with real, messy desktops.

Where Terra fits — and where it doesn't

• Use Terra for: high-volume computer work — invoice entry, form filling, document processing, proposal drafting, reading technical drawings, matching photos to catalogue entries, time-coded video logs.
• Keep Sol or Astra for: the hardest coding and research. On Terminal-Bench, Terra's 50.3% trails Sol's 55.9% and Claude Sonnet 5.5's 70.6%.
• Keep Luna for: the simplest, cheapest tasks. At $0.10 input and $0.50 output, Luna is still roughly a tenth of Terra's price for work that does not need visual reasoning or computer use.

What to check before you switch

• Audit what you run on Sol or GPT-5.6 Terra today. Any agent workload that is screen-bound, form-bound or document-bound — not hard-coding-bound — is a candidate to move to gpt-6.1-terra.
• Watch the 272K cliff. If your prompts regularly cross 272,000 tokens, chunk them or budget at the higher rate.
• Do not give a computer-use model unsupervised write access on day one. Start it on read-only or review-queued work, check what it actually did, and only then widen what it is allowed to change.
• Wait for independent benchmarks before rewriting your model strategy. One independent run on your own workload beats ten launch charts.

FAQ