GPT-6.1 Terra at work

OpenAI has released GPT-6.1 Terra, a faster, cheaper model built for the everyday computer tasks that fill a workday — and priced it at half the cost of the model it replaces.

In its announcement, OpenAI describes Terra as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing.” The pitch is aimed squarely at routine work: drafting proposals, interpreting technical drawings, filling out forms, editing videos. Its example for developers is telling — automating high-volume chores “like reading incoming invoices and entering their details into accounting software.”

The price cut

Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens — 50% below both GPT-5.6 Terra and GPT-6.1 Sol. The model is available now in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and to developers through the API as gpt-6.1-terra. OpenAI says the same plan allowance covers more tasks with Terra than with its bigger models.

The cut is not even across the board, and the details favor heavy agent use: output tokens fall 58% from GPT-5.6 Terra’s $12, and cached input drops 75%, from $0.20 to $0.05. For agents that run continuously in the background — re-reading context, retrying steps, checking their own work — that cached-input price is arguably the more important number.

What OpenAI claims it can do

All of the performance figures below are OpenAI’s own, from its announcement post, and should be read that way:

  • Agents’ Last Exam (a computer-use benchmark): 15 points higher than GPT-5.6 Terra at medium effort, at about one-third the cost.
  • Terminal-Bench 4.0 (command-line coding): 50.3% of tasks solved at maximum effort, up from 21.5%.
  • Error-inducing prompts (an internal set designed to trip models up): error rate fell from 53.6% to 16.7% at low reasoning effort.
  • Speed: over three times faster than GPT-6 Astra and GPT-6.1 Sol on task completion.

OpenAI also claims Terra matches GPT-6 Astra on some agent tasks at one-tenth the token price, and says it is “over three times as token efficient as Claude Opus 5.5” at maximum effort — again, its own measurement on its chosen benchmark. Independent testing will tell us how much of that survives contact with real workflows.

The model carries a 1.05-million-token context window and, notably, a safety footnote: OpenAI says Terra reaches the Critical cybersecurity threshold under its Preparedness Framework, meaning it can identify and exploit previously unknown software flaws. The company says GPT-6 Astra protections apply, and a system-card addendum accompanies the release, reporting closer scope adherence, more honest disclosure of limitations, and resistance to prompt injection in tests.

Why it matters

Terra is not a flagship, and OpenAI is upfront that Sol and Astra remain the better picks for the hardest coding jobs. That is precisely the point. The flagship race gets the headlines, but the mid-tier price war is where most developers’ bills are decided — and Terra pushes that floor lower still, from the open-weight side as well as the closed side, with DeepSeek’s latest coding model squeezing the same tier.

The ecosystem is already moving: Microsoft is integrating Terra into its Foundry platform, and GitHub Copilot is evaluating it for coding workloads. If you already use GPT-5.6 Terra, the switch is straightforward — same tier, half the price, better scores on OpenAI’s tests. If you use a competitor’s model for agent-style work, Terra is now the cheaper default to test against, though you should benchmark it on your own tasks before migrating.

The broader signal is about where AI value is shifting. The models are no longer competing mainly on what they know; they are competing on what they can do inside the software people already use, and at what price per task. Terra’s bet is that computer use — the unglamorous work of forms, invoices, and desktop apps — is the next battleground, and that winning it means being fast and cheap enough to run all day. The price cut is real money from day one. Whether the capability holds up is the question to watch.

Sources: OpenAI — “Introducing GPT-6.1 Terra” (Oct 26, 2026); Digital Soul Hub — “GPT-6.1 Terra: OpenAI’s cheaper computer-use model”; Testing Catalog — “OpenAI launches GPT-6.1 Terra for precise computer use”; The AI News Report — “GPT-6.1 Terra pricing explained”. All benchmark figures are OpenAI’s own and await independent verification.