GPT-6.1 Terra: What It Is and When to Switch

GPT-6.1 Terra is the model to evaluate first when your work involves understanding images, operating apps, or writing clearly from supplied material. OpenAI's October 26, 2026 release puts those jobs at half GPT-6.1 Sol's standard API token rates. Keep Sol for complex projects, GPT-6 Astra for the hardest work, and GPT-6 Luna for simple repeatable tasks.

Verified against the makers' live documentation on October 27, 2026. The switching recommendations are analysis of those sources; the cost example is calculated, not a measured workload.

What Is GPT-6.1 Terra?

GPT-6.1 Terra is OpenAI's fast, efficient model for visual reasoning, computer use, and natural writing. Visual reasoning means interpreting what an image shows, including relationships and layout. Computer use means taking actions through software interfaces. OpenAI describes the release as an upgrade from GPT-5.6 Terra for everyday professional work, including proposals, technical drawings, and forms. OpenAI's announcement

Think of the difference between reading a map and following it. An image-capable assistant can describe a screen. A computer-using agent also needs to select an action, inspect the changed screen, and establish whether it reached the intended result. Terra's intended niche combines that visual understanding with action.

For an operator, that suggests a bounded assignment such as matching supplier photos to existing catalog entries and preparing listing updates. For a founder, it suggests turning supplied product notes into a proposal. These are candidate workflows, not promises that an unattended run will get every record or sentence right.

The model also needs an environment that supplies the tools and permissions. Selecting Terra does not itself connect an accounting system or grant access to a browser. OpenAI's Work documentation explicitly separates model availability from browser and computer-use setup. Work release notes

GPT-6.1 Terra vs Sol: Which Work Should Move?

Move defined visual and writing tasks to a Terra trial; keep difficult, interconnected projects on Sol until the trial earns a change. OpenAI's model-selection guidance recommends comparing the same inputs at the same reasoning effort, the setting that controls how much deliberation a model uses. Its recommendation for Sol remains complex work where time and cost matter. OpenAI's selection guidance

Specifications: Terra model card and Sol model card. A context window is the working token budget, not a guarantee that every detail will be used correctly; the maximum output allowance is not extra space on top.

Builders: a solo technical builder can trial Terra on a bug with a clear reproduction and a regression test. Keep Sol on a change spanning authentication, billing, and data migration, where each decision affects the others. This is a practical application of OpenAI's distinction between scoped fixes and complex coding, not a claim that Terra cannot attempt the larger job.

Operators: a senior operator can trial Terra on filling known fields from a supplied document. The acceptance check is the saved record, including fields that should remain untouched. A confident completion message alone does not establish success.

Buyers: a mid-market CTO should change the model assignment for a workflow before changing the whole company's default. A funded founder can move a source-based proposal draft into the trial while keeping a strategy project with conflicting evidence on Sol or Astra. Compare review and repair time alongside speed.

When Astra or Luna Still Makes More Sense

GPT-6 Astra remains the escalation for demanding reasoning and ambiguous work; GPT-6 Luna remains the economical starting point for focused, high-volume tasks. Those are OpenAI's published roles. A larger version number on Terra does not put it above every member of the GPT-6 family. OpenAI's GPT-6 guide

If Luna already classifies an incoming request or extracts a few fields accurately, leave that job there. Try Terra when visual interpretation, interaction with apps, or writing quality is the reason the cheaper workflow fails. Keep Astra for a difficult investigation or deliverable where the quality difference can justify more usage. The rule is to choose the least expensive option that passes your actual acceptance checks, not the cheapest attempt.

For the broader model family and exact identifiers, use the site's GPT models guide.

Should You Switch From GPT-5.6 Terra?

Start a migration trial now if GPT-5.6 Terra handles your visual or everyday app work. OpenAI's migration guidance says to compare representative images and workflows at the same reasoning effort. Start at medium, then change effort only after establishing a baseline. Applications using Chat Completions for tool calls must move those requests to Responses when adopting GPT-6.1 Terra. Migration guidance

The older GPT-5.6 family guide describes the previous lineup. The case for trying the new Terra is stronger than the case for blindly replacing it. Keep the old path available while checking documents with unusual layouts, ambiguous product photos, incomplete inputs, and tool failures. These are examples of evaluation cases to choose, not failures observed in this run.

What OpenAI's Benchmarks Establish

The published results support a trial, with clear limits on what they predict. All results below are vendor-reported by OpenAI, not independent tests:

• On Agents' Last Exam, Terra at medium effort scores 15 percentage points above GPT-5.6 Terra at about one-third the cost. OpenAI reports task completion over three times faster than Sol and Astra in this evaluation.
• On MMMU-Pro visual reasoning, OpenAI says Terra nearly matches Sol and Astra at lower cost.
• On Terminal-Bench 4.0, Terra at maximum effort solves 50.3% of tasks versus GPT-5.6 Terra's published 21.5%. OpenAI still recommends Sol or Astra for the most complex coding.

Source: OpenAI's launch benchmarks

A benchmark fixes a task set and evaluation setup. Your workflow adds different prompts, permissions, tools, and failure costs. A result on desktop tasks does not establish that Terra replaces Sol for every research or engineering problem.

A Worked Switching Budget

At a fixed token mix, Terra halves Sol's standard model-token bill, while Luna remains cheaper. OpenAI's standard short-context rates per million tokens are $1 input, $0.05 cached input, and $5 output for gpt-6.1-terra. The corresponding input/output rates are $2/$10 for Sol, $10/$50 for Astra, and $0.10/$0.50 for Luna. OpenAI API pricing

This calculation assumes no cache writes or reads, separate tool charges, speed or regional premiums, and no request above the long-context threshold. Terra's model card applies higher full-request rates above 272K input tokens, so a million-token context window does not mean million-token requests retain the base rate. Terra pricing conditions

Actual models can consume different amounts of reasoning and tool context, and retries can change the total. Measure cost per accepted result before treating these token savings as workflow savings. The full billing explanation belongs in the site's OpenAI API pricing guide.

Where Does GPT-6.1 Terra Run?

Terra is available through OpenAI, Amazon Bedrock, and GitHub Copilot; Microsoft's own documentation also lists it in Azure's Foundry model offering. Choose the route your organization can administer and evaluate, then check that route's access conditions.

ChatGPT Work and Codex Plans

ChatGPT Work and Codex list Terra on Plus, Pro, Business, Enterprise, and Edu plans. The current Work release notes explicitly include Business, although the launch announcement's list omits it. Choose Terra from Advanced in the desktop app or Work on the web; Work also supports it on mobile. It is not available in regular Chat. Enterprise access depends on workspace and administrator settings, and the release notes exclude UAE inference residency. OpenAI's Work documentation

For a buyer, this separates entitlement from capacity: a plan can include a model without promising a fixed number of completed jobs. OpenAI says Terra makes the existing allowance go further, but publishes no universal task multiplier in the announcement. API price ratios do not establish subscription allowance ratios. See Codex pricing and plan access for the subscription side.

OpenAI API

Developers select gpt-6.1-terra; use the Responses API when the workflow calls tools. The model accepts text and image inputs and produces text. Its model card does not list native audio or video support. A model helping operate a video-editing application is different from an API accepting video as an input modality. Terra API documentation

Amazon Bedrock

Amazon Bedrock, AWS's managed model service, announced Terra's general availability on October 26, 2026. AWS says you can call it through Bedrock APIs or configure ChatGPT Work and Codex to use Terra on Bedrock. Its announcement directs customers to Bedrock documentation for regions, endpoints, supported features, inference profiles, and pricing. AWS's announcement

For an AWS-based operator, this provides a route to evaluate under the existing AWS access and audit setup. It does not establish identical features or billing across every provider.

GitHub Copilot

GitHub Copilot, GitHub's coding assistant, announced Terra's general availability on October 26 for Pro, Pro+, Max, Business, and Enterprise. GitHub lists VS Code, Visual Studio, Copilot CLI, its cloud agent and app, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse. Rollout is gradual; Business and Enterprise administrators can control access through model policy. GitHub's announcement

GitHub says the model uses provider list pricing under usage-based billing. Evaluate it in the client your developers already use, and confirm the model policy if it has not appeared in their picker.

Microsoft Azure

Microsoft Azure's Foundry reasoning-model documentation explicitly lists gpt-6.1-terra, version 2026-10-26. It includes Terra in the availability and feature tables and points to region availability and quota tiers. That supports including Azure here; it does not establish availability in every region or deployment type. Microsoft's model documentation

What's Overhyped?

The overreach is treating a cheaper model with strong visual results as a universal replacement. Terra's place depends on the job, the tools around it, and the quality of the saved outcome.

A faster benchmark run does not make every production task faster. A lower token rate does not mean the same reduction in a subscription's usage meter. And availability in a model picker does not mean an agent can access every app it needs.

The limitation that matters most for operators is verification: a model can describe a plausible completed task without the intended record being saved correctly. OpenAI's selection guidance specifically recommends reopening affected records, checking unrelated data, and comparing the completion report with the result. Build that check into the trial, not just a subjective rating of how convincing the response sounded. OpenAI's evaluation criteria

The Monday Move: Change One Workflow

Pick one recurring workflow with a checkable result and compare Terra with its current model. Use the same source material, permissions, and reasoning effort so the comparison answers a useful question.

1. 1

   Define a pass before running the task

   For catalog work, require the correct matched product, the requested saved fields, and no unrelated edits. For a proposal, require the supplied facts and the right audience and voice. For code, specify the behavior and regression checks.

2. 2

   Compare the complete outcome

   Record completion time, billed usage, review effort, and repairs. Reopen saved records or inspect the deliverable. Do not score only the final chat message.

3. 3

   Promote the workflow that passes

   Move it to Terra if it meets the quality bar and improves the total cost or time. Leave reliable simple work on Luna, retain Sol for complex projects, and escalate the hardest work to Astra. Revisit the choice when the workflow changes.

Subscribe to the newsletter for practical guidance on model choices and business workflows.