GPT-6.1 Terra Is Here. Should You Choose It Over Sol?

A client brief does not always need your most expensive AI model. Neither does turning meeting notes into a proposal or pulling the important changes out of a dashboard. The cheaper option earns its place when it finishes that work well enough to save you time.

GPT-6.1 Terra, introduced on October 26, is OpenAI's latest attempt to make that trade-off attractive. It combines writing, visual understanding and computer use at lower API rates than GPT-6.1 Sol. For people doing recurring professional work, the appeal is straightforward: more useful output from the budget they already have.

There is one distinction to understand before hunting through a model selector. Terra is available in supported ChatGPT Work and Codex experiences, not ordinary ChatGPT Chat, according to the October 26 changelog. Our ChatGPT and OpenAI Codex profiles explain the products that contain those workflows.

Our early verdict: Terra is worth trying for writing and visual tasks with a clear finish line. Keep Sol available for complexity. Independent results show a substantial upgrade over the previous Terra, while also showing why the cheapest token price does not always produce the cheapest completed job.

How we assessed this: We reviewed OpenAI's documentation and Artificial Analysis's independent results. Choosely has conducted no hands-on model tests; the suggested trials below are editorial recommendations.

What Terra could take off your plate

OpenAI's model guide recommends Terra for everyday writing and visual work, including proposals, reviewing visual material and completing forms from source documents.

Its dashboard example makes the idea concrete. Give Terra screenshots and ask it to explain the main changes, grounding its summary in the chart labels and values and identifying anything it cannot read confidently. That could help turn a busy reporting screen into the first draft of a client update.

The useful result would preserve the figures, distinguish an actual trend from a change in the chart's scale, and leave you with a short explanation you can verify. You would still check it against the dashboard before sending it.

A proposal from supplied notes offers a similar starting point. The source material and requested outcome give the model something definite to work toward. A recurring task also gives you a baseline: you know how long it usually takes and which mistakes would make the output unusable.

This is where a less expensive model can become valuable. Routine preparation fills a surprising amount of the working week. Saving the stronger model for the parts that need deeper judgment could make the whole workflow more economical.

Who gets it, and where to look

The launch changelog lists Plus, Pro, Business, Enterprise and Edu for Terra in Work and Codex. Business is included there despite its omission from the launch announcement's plan list. Free and Go have no announced Terra entitlement.

On the desktop app and Work on the web, open Advanced to select Terra. Work also supports it on mobile. Workspace permissions and rollout conditions can affect what appears, so managed-account users should check with their administrator if it is missing.

For command-line use, update to Codex CLI 0.159.0 or later and select codex -m gpt-6.1-terra. The model guide's Terra entry omits the IDE-extension availability badge shown for the other models; IDE access remains unconfirmed in this assessment.

The changelog also excludes UAE inference residency, a specific processing-location configuration. Organizations using that configuration should check their deployment options before planning a switch.

Developers have a separate route through the API, with separate billing. Your ChatGPT plan doesn't include API usage.

Terra got much better. Sol still leads.

Artificial Analysis found a six-point gain over GPT-5.6 Terra on its Intelligence Index at maximum effort. Its estimated benchmark-task cost fell from $1.40 to $0.40. On the newly released Coding Agent Index v2.0, Terra scored 48 against Sol's 54.

Effort settings matter here. They control how much reasoning work the model does, which affects quality, time and usage. Artificial Analysis places Terra at low and medium effort on its intelligence-versus-cost efficiency frontier, among the best measured balances of capability and cost. Maximum effort sits behind that frontier.

Start with the default setting. The API default is medium; app defaults vary by account, so compare medium where your interface offers it. Turning every control up can spend away the advantage that attracted you to Terra.

The higher-effort comparison reveals another useful wrinkle:

Artificial Analysis Intelligence Index v4.3.2, checked October 27. Scores are rounded aggregates; costs are weighted benchmark estimates in USD. Your workflow and subscription usage may differ.

Sol at medium effort reaches the same rounded overall score as Terra at maximum, for a lower estimated task cost. This compares two different configurations. Individual strengths can differ even when the aggregate score matches.

For the person choosing a model, it is a reminder to count the whole job. Include generated material, repeat attempts and corrections when comparing the total cost.

What subscribers gain

OpenAI says Terra lets your included allowance go further. Its developer announcement attaches no universal task count or guaranteed multiplier to that promise.

The practical opportunity is to try it on work you repeat. If the finished result holds up and you spend less time correcting it, you have found a useful role for Terra. A failed first attempt followed by a long repair conversation belongs in that assessment too.

You also have room to choose within a project: Terra for a draft from clear notes, Sol for resolving contradictions across sources, and Astra when the task calls for difficult judgment.

When to use Terra, Sol or Astra

Try Terra when the inputs and finish line are clear. Start with the dashboard summary, a proposal from supplied notes or a form populated from a source document. Use the default effort first, then compare medium if your interface offers it. For a bounded coding trial, our suggestion is a small change with a test that tells you whether it worked.

Review the actual deliverable. Check figures, names, omitted requirements and saved changes. Keep approvals enabled for consequential actions and grant only the permissions the task needs. OpenAI's system card reports weaknesses in respecting restrictions and authorization boundaries in controlled evaluations, including a warnings test run without safeguards. Those conditions limit what the results tell us about everyday use, while reinforcing the value of clear permissions.

Use Sol when the pieces interact. Larger code changes, conflicting source material and complex professional deliverables can require sustained reasoning across several decisions. OpenAI's announcement directs complex coding and in-depth research toward Sol or Astra. If Terra repeatedly misses dependencies or needs extensive repair, compare Sol at a moderate effort setting before spending more on Terra's maximum.

Use Astra when the difficult part is the judgment. OpenAI positions it for its most demanding work, including ambiguous problems and complex end-to-end tasks. A decision involving competing requirements may justify that extra capability and expense. Our Astra introduction explains its broader role.

For simpler, high-volume tasks, Luna remains another option at lower API rates. Model choice can stay flexible as the work changes.

Existing GPT-5.6 Terra developers also have time to evaluate a switch: the deprecation schedule checked for this article lists no retirement deadline for that model.

Choosely's take

Terra's strongest case is everyday work you can describe clearly and inspect afterward. The independent evidence supports a meaningful upgrade, and its lower-effort settings deserve attention before anyone concludes that Sol is always the better value.

Our recommendation is to try Terra on one recurring task at the default or medium setting, judge the finished result, and keep the stronger models available for the parts it struggles with. Include the corrections and review time in your decision. A cheaper job that is properly finished is a useful upgrade.

If ChatGPT or Codex is part of your regular workflow, save your stack with Choosely. Keeping the tools you use together gives you a place to review them as access, pricing and capabilities change.