GPT-6.1 Terra: Pricing, Access and Terra vs Sol

OpenAI released GPT-6.1 Terra on October 26, 2026. It is a fast model for professional work. Paid ChatGPT plans get it in ChatGPT Work and Codex, and developers get it in the API. The API price is $1 per million input tokens and $5 per million output tokens, half of GPT-6.1 Sol. By the end you will know who can use GPT-6.1 Terra and what it costs. You will also see how it compares with Sol and the older GPT-5.6 Terra, and when we would pick it.

What is GPT-6.1 Terra, in one sentence?

GPT-6.1 Terra is a fast GPT-6 series model from OpenAI, priced at half of GPT-6.1 Sol and built for writing and computer use with strong image understanding.

The Codex changelog entry for October 26 calls it "a fast, efficient model for professional work." OpenAI's system card addendum, published the same day, aims it at "high-volume professional work." The Codex models page now lists four GPT-6 series models, the fourth being GPT-6 Luna. GPT-6 Astra is the most capable one. Terra sits below Sol on price. Its API model ID is gpt-6.1-terra.

How does GPT-6.1 Terra differ from Sol and the older Terra?

The closest comparisons are GPT-6.1 Sol and GPT-5.6 Terra, the previous Terra. Sol has been the Codex CLI's default model since October 7, per the changelog.

Sources: OpenAI's API model pages, linked in the column headings.

Against Sol, Terra costs exactly half on every line. Compared with GPT-5.6 Terra, input is half price. Output falls from $12 to $5 per million tokens. Terra also has the newest knowledge cutoff of the four.

Who can use GPT-6.1 Terra, and where?

The changelog lists Plus, Pro, Business, Enterprise and Edu plans. Enterprise access depends on workspace availability and admin controls. Two limits matter:

• Terra is not available in ChatGPT Chat. You get it in ChatGPT Work and Codex only.
• It does not support UAE inference residency.

In the ChatGPT desktop app you pick it from the model list under Advanced. It is also in ChatGPT Work on the web and on mobile. In the Codex CLI, Terra arrived in version 0.165.0 on October 27. The models page gives the command as codex -m gpt-6.1-terra.

In the API, the same model ID works with the Responses API and Chat Completions. Batch works too. The model takes text and images and returns text, with no audio or video support and no fine-tuning.

What does GPT-6.1 Terra cost?

In ChatGPT, OpenAI's changelog lists no separate price for Terra. We found no OpenAI page we could open that says how Terra usage counts against plan limits compared with Sol.

In the API, the model page lists $1 per million input tokens and $5 per million output tokens. Cached input costs $0.05 per million. A few modifiers change the bill:

• Prompts over 272K input tokens cost 2x on input and 1.5x on output for the whole request.
• Batch and Flex processing are 50% cheaper than standard.
• Fast mode costs 2x standard.
• Regional processing adds 10% where it is offered.

Say an agent turn reads 100K tokens and writes 5K. If 80K of the input is cached, the turn costs about $0.05 on Terra. The same turn on Sol is about $0.10, and on Astra about $0.53. Long agent sessions are mostly cached input, so the cached rate matters more than the headline price.

When should you pick Terra over Sol?

OpenAI's Codex models page says to choose Terra for "clear communication" and for "interpreting images and screens." It also suggests Terra for finishing tasks across supported apps. For repeated, long-running work, it points to Sol. Astra stays the pick for the hardest work.

Our reading of that guidance:

• Pick Terra for drafting and screenshot reading, plus high-volume API jobs where cost per call decides the design.
• Pick Sol or Astra for deep refactors and long debugging or research sessions, where a wrong step costs more than the tokens.

Terra also has one API option Sol lacks. Its reasoning effort goes from none up to max, with medium as the default. Sol starts at low. A "none" setting is useful for cheap classification or extraction calls inside an agent loop.

All four models share the same 1,050,000-token API window. Inside Codex and ChatGPT Work, the models page says Sol defaults to a 500,000-token window, and it does not state a Terra default.

What should you read in the fine print?

Almost everything published about Terra so far, including its positioning and comparisons, comes from OpenAI. We found no independent benchmark of Terra with a published setup.

The system card addendum publishes some of OpenAI's own evaluation results, run on its own setups. Several place Terra below Sol. On Sandbox Bench, a cyber evaluation, Terra reached 5 of 22 targets. Sol reached 9 and Astra 10. On HealthBench Professional, OpenAI reports 62.7 for Terra. GPT-5.6 Terra scored 57.7 and Sol 64.2. All of these numbers are self-reported by OpenAI.

OpenAI treats Terra as Critical in cybersecurity and High in biological and chemical capability under its Preparedness Framework. It applies the same safeguards stack it uses for Sol and Astra.

Secondary sites also report a Terminal-Bench 4.0 score and a speed multiple for Terra, credited to OpenAI's announcement. We could not open that page, so we left those numbers out. One more gap: OpenAI's computer use docs still recommend Astra for difficult tasks that depend on visuals. They do not mention Terra yet.

The bottom line

Pay-per-token users should try Terra first on drafting and extraction jobs. Keep Sol for coding work you cannot afford to redo. In Codex, switch to Terra for routine tasks and switch back for hard ones. Test it on your own tasks before you change a default, because the only published evaluations are OpenAI's.

How do you keep agent context when you switch models?

Following the advice above, you plan a refactor with Sol in a Codex thread. Then you pick Terra from the model list to draft the release notes. In the afternoon you open the same repo in Claude Code to finish the work. Claude Code has none of the morning's plan, and you explain it again.

Codex and ChatGPT Work let you swap models inside OpenAI's tools. memU keeps one readable memory across tools from different vendors. Codex and ChatGPT Work mode are among them.

Inside Codex and ChatGPT Work, switching between Sol and Terra is a pick from the model list. If you never leave those two, that needs no extra setup. memU is worth adding when Claude Code or Cursor joins the workflow, or when you want the remembered facts as files you can read and fix.

The hosted version on memu.so is free and syncs across devices. To try it, get an API key from memu.so and send your agent one message:

Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU, API Key is YOUR_MEMU_API_KEY.

The self-hosted setup and the support matrix for each host are in the memU GitHub repository, and the product overview is at memu.pro.

FAQ

How much does GPT-6.1 Terra cost in the API?

OpenAI lists $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Prompts over 272K input tokens cost more. Batch and Flex are 50% cheaper.

Can you use GPT-6.1 Terra in ChatGPT chat mode?

No. OpenAI's changelog says Terra runs only in ChatGPT Work and Codex on paid plans. It is not available in Chat.

Is GPT-6.1 Terra better than GPT-6.1 Sol?

Not on OpenAI's own system card evaluations, where Terra scores below Sol on several tests. It costs half as much in the API. OpenAI positions Sol for complex, long-running work.

How do I use GPT-6.1 Terra in Codex?

Pick it from the model list in the ChatGPT desktop app, or run codex -m gpt-6.1-terra in the Codex CLI. Terra arrived in Codex CLI 0.165.0 on October 27.

What is the GPT-6.1 Terra context window?

The API page lists 1,050,000 tokens, with up to 922,000 input tokens and 128,000 output tokens. The Codex models page does not state a separate default for Terra inside Codex or ChatGPT Work.