GPT-6.1 Terra review

OpenAI's new mid-price model for computer use and everyday work, at half the price of GPT-6.1 Sol.

Out nowOpenAIAI modelUpdated Oct 26, 2026
Try GPT-6.1 TerraOpen in ChatGPT · official source, chatgpt.com
Released26 Oct 2026
Price (API)$1 in / $5 out
Best forComputer use, everyday work
First lookOur take, based on published facts and early independent tests

The new sweet spot in OpenAI's line-up. It costs half as much as GPT-6.1 Sol, and independent tests from Artificial Analysis put it only a few points behind Sol. It is a big jump over the old GPT-5.6 Terra, especially on computer use and on not making things up. For the hardest coding work, OpenAI itself still points you to Sol or Astra.

What is GPT-6.1 Terra?

GPT-6.1 Terra is an OpenAI model released on 26 October 2026. It is the mid-price tier of OpenAI's four models, Luna, Terra, Sol and Astra, and replaces GPT-5.6 Terra, which the GPT-6 launch had skipped. OpenAI built it for using a computer, understanding screenshots and images, writing and everyday coding.

Key features

  • Half the price of GPT-6.1 Sol: $1 per million input tokens and $5 per million output tokens
  • OpenAI says it finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol
  • Far fewer made-up answers: Artificial Analysis measured its hallucination rate falling from 88% to 38% against GPT-5.6 Terra
  • Reads about 1 million tokens at once and writes up to 128,000
  • In ChatGPT Work and Codex on paid plans, the OpenAI API and OpenRouter

Specs and price

Made byOpenAI
Released26 Oct 2026
API price$1 per 1M input tokens, $5 per 1M output tokens
Repeated input (cache)$0.05 per 1M tokens
How much it can read1,050,000 tokens
Longest answer128,000 tokens
ReadsText, images and files
WhereChatGPT Work and Codex on Plus, Pro, Enterprise and Edu. Free plans are not mentioned
Free to downloadNo
Name in the APIgpt-6.1-terra

Test scores

TestScorePublished by
Terminal-Bench 4.0Coding in a terminal, max effort50.3% (GPT-5.6 Terra: 21.5%)OpenAI
OpenAI error testWrong answers on tricky prompts, max effort4.3%, close to GPT-6 Astra and GPT-6.1 SolOpenAI
Coding Agent Index v2.0Independent coding-agent score48 (GPT-6.1 Sol: 54, GPT-6 Luna: 31)Independent
AA-OmniscienceHow often it guesses instead of saying it does not know38% (GPT-5.6 Terra: 88%)Independent
Cost per taskArtificial Analysis, max effort$0.40 (GPT-5.6 Terra: $1.40)Independent

Scores published by OpenAI are the maker's own numbers. We will add our own results in the hands-on review. Independent scores are from Artificial Analysis.

How to use GPT-6.1 Terra

  1. In ChatGPT: sign in on a Plus, Pro, Enterprise or Edu plan and pick GPT-6.1 Terra in ChatGPT Work or Codex.
  2. For developers: use gpt-6.1-terra in the OpenAI API.
  3. Also available on OpenRouter as openai/gpt-6.1-terra.

Who it is for

Good fit

  • Developers who want near-Sol quality at half the price
  • Agents that click through websites and apps
  • Paid ChatGPT users who do everyday writing and coding in Codex

Skip it if

  • The hardest coding projects: OpenAI recommends GPT-6.1 Sol or GPT-6 Astra
  • Free ChatGPT users: OpenAI's launch only lists paid plans

Pros and cons

Pros

  • $1 / $5 per million tokens, half of GPT-6.1 Sol
  • Much fewer made-up answers than the last Terra
  • Huge 1M token memory
  • Independent tests back up most of the jump

Cons

  • A few points behind GPT-6.1 Sol in independent tests
  • Speed and Astra comparisons come from OpenAI
  • Not listed for free ChatGPT plans

How it compares

GPT-6.1 Terra costs half of GPT-6.1 Sol ($2 / $10) and ten times GPT-6 Luna ($0.10 / $0.50). Artificial Analysis puts it 4 points below Sol and 10 points above Luna on its Intelligence Index, so for most work it is the cheaper choice with little loss.

Hands-on review In testing

We are running GPT-6.1 Terra on real tasks. Results will be added to this page, at the same address.

Real tasksCoding, writing, research, agents
SpeedHow fast it answers in practice
Cost per taskWhat a typical job costs
VerdictWho should switch, who should wait

Related models