GPT-6.1 Terra vs Sol: benchmarks and pricing
Terra 6.1 targets visual work and clear writing. Explore published benchmarks, Sol comparisons, API pricing and access in Work, Codex and Rankerfox.
GPT-6.1 Terra returns with a visual focus
GPT-6.1 Terra launched on October 26, 2026, according to the official OpenAI changelog. Following Sol 6.1, it focuses on images, on-screen actions and clear writing. It succeeds GPT-5.6 Terra, so the Terra name is no longer limited to the previous generation.
For professional work, the point is not to collect model names. It is to find a useful balance for frequent tasks: reading a screenshot, matching a product to a reference, revising a page or making a precise change. Our recommendation: try Terra on well-scoped work and compare Sol on more interconnected projects.
Terra or Sol: what is different?
OpenAI's model selection guide points to Terra for concise writing, visual tasks and actions with a clearly defined outcome. Sol remains a starting point for complex projects, coding and analysis across sources. These are useful starting points, not hard boundaries: Sol can also work with images and use a computer.
| Criterion | GPT-6.1 Terra | GPT-6.1 Sol |
|---|---|---|
| Starting point | Visual tasks, direct writing and well-scoped actions | Coding and complex, multi-step projects |
| Input / output | Text and images / text | Text and images / text |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Maximum output | 128,000 tokens | 128,000 tokens |
| API reasoning efforts | none, low, medium, high, xhigh, max | low, medium, high, xhigh, max |
| Default API effort | medium | medium |
Technical details appear on the Terra 6.1 and Sol 6.1 model pages. A large context window lets you supply more material; it does not guarantee that every detail will be retrieved. Audio and video are not native Terra modalities, although an application may use separate tools for those functions.
Early benchmarks: where does Terra 6.1 stand?
In its October 26 report, Artificial Analysis gives these results for its Coding Agent Index v2.0:
| Model | Published index |
|---|---|
| GPT-6.1 Sol | 54 |
| GPT-6.1 Terra | 48 |
| GPT-6 Luna | 31 |
Terra sits between Sol and Luna on this index. 48 does not mean a 48% success rate. The post does not specify effort separately for each row. These are external evaluations, not Rankerfox tests or SEO measurements. The v2.0 protocol differs from the historical snapshot in our Sol article, so scores cannot be compared directly across the two tables.
Four practical SEO and marketing uses
Read a screenshot without overinterpreting it
For a Search Console screenshot, first ask for the visible period, filters, units and changes. A decline for one query does not necessarily mean the entire site declined. For complete calculations and comparisons, provide the export instead: missing information cannot be reliably reconstructed from an image.
Review campaign assets
A clipped logo, text too close to an edge or a thumbnail that becomes unreadable on a phone makes a useful visual review task. Specify the final format and the elements that must remain visible. For a catalog, match photos to the references actually provided rather than asking the model to invent a missing product reference.
Rewrite a page from precise facts
Supply a product sheet, approved prices and restrictions, then request a title, introduction or FAQ. Check whether the facts survived, not just whether the copy reads smoothly. Persuasive wording must not turn a limited allowance into unlimited access.
Make a small, verifiable change
Editing a CMS field or a few lines of code is a better first test than an entire redesign. Define the expected result and permitted actions. Afterward, verify the saved page or passing test, rather than relying solely on a message saying the work is complete.
API pricing: half Sol's per-token rates
For the three text-token categories below, Terra's Standard rates are 50% lower than Sol 6.1's. That does not guarantee half the bill for an entire task.
| Text tokens | Terra 6.1 | Sol 6.1 |
|---|---|---|
| Uncached input | $1.00 | $2.00 |
| Cached input | $0.05 | $0.10 |
| Output | $5.00 | $10.00 |
Calculated example: 20,000 uncached input tokens and 2,000 billed output tokens cost $0.03 with Terra or $0.06 with Sol, excluding tools and surcharges. With identical token volumes, the ratio is straightforward; in actual use, reasoning and retries can change consumption.
Above 272,000 input tokens, both model pages specify twice the input/cache rates and 1.5 times output rates for the full request. Cache writes and tools have separate charges. API prices are neither a subscription price nor a Rankerfox allowance.
How to compare Terra and Sol on your own work
Choose a few representative tasks, not just the one that makes the best demonstration. Give both models the same documents and instructions; keep the same effort where both support it, such as medium. Then record:
- Accuracy: are facts, calculations and references preserved?
- A verifiable result: does the file, page or change actually exist in the expected state?
- Rework: how many follow-ups and human corrections are needed?
- Total cost: working time and, for the API, actual usage including tools.
Keep Terra when it meets your standard with less effort. Move to Sol when dependencies between steps or documents require more continuity. Repeated factual errors matter more than a pleasantly written response.
Availability in Work, Codex and Rankerfox
OpenAI announces Terra 6.1 in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans, subject to rollout and workspace controls. In the desktop app and Work on the web, choose it in the model list under Advanced. Work also supports it on mobile; it is not available in regular Chat.
ChatGPT Pro is part of Rankerfox Premium, alongside the other AI tools. For Terra 6.1, check the model name in the Work or Codex picker of the access you are using: ChatGPT Pro access alone does not confirm this particular model. See Rankerfox plans for the contents of each offer.
Frequently asked questions
Does Terra 6.1 replace Sol 6.1?
No. Terra offers another balance of cost and use cases. Our Sol 6.1 guide remains useful for complex work and for understanding the evaluations published at its launch.
Can it generate an image directly?
It accepts images as input, but its native output is text. Image generation through a tool is a separate function: understanding an image and producing one are different capabilities.
Do I need max effort for a good result?
Not by default. Start with an intermediate setting and measure the corrections required. Higher effort should earn its place through a demonstrated improvement on the task, not its name.
Sources and reference date
Information checked on October 28, 2026. Independent results refer to the October 26 article; prices are Standard API rates, not subscription prices.