The short answer: Terra for volume, Sol for hard engineering
Choose GPT-6.1 Terra if you run many routine tasks such as reading invoices, filling forms or drafting text, since it lists $1 input and $5 output per million tokens and has a none effort setting. Choose GPT-6.1 Sol for complex coding and multi-step agent work: OpenAI points to Sol or Astra for the most complex coding, at $2 and $10 per million tokens.
Scope and evidence
Task: picking one of OpenAI's two GPT-6.1 models for API work or for ChatGPT Work and Codex. Prices are in US dollars from OpenAI's model pages, the Terra announcement and the Work and Codex usage guide, plus cost per task from Artificial Analysis Intelligence Index v5, all checked on October 28, 2026.
AppFor.ai ran neither model for this page. Artificial Analysis is an independent benchmark; OpenAI's launch figures are vendor claims.
Compare the same criteria
Each line sets the two GPT-6.1 models against each other from OpenAI's pages, then says what the gap means.
- Positioning. Terra: visual reasoning, precise computer use and natural writing. Sol: complex coding, computer use and professional work. What this means: Terra is the volume model, Sol the engineering model.
- API price per million tokens. Terra: $1 input, $0.05 cached, $5 output. Sol: $2 input, $0.10 cached, $10 output. What this means: Terra is exactly half the price on every line.
- Reasoning effort. Terra: none, low, medium, high, xhigh and max, default medium. Sol: low to max, default medium. What this means: only Terra can skip reasoning entirely.
- Context and output. Both: 1.05M-token context, 922K max input, 128K output, text and image input. What this means: no difference for long documents.
- Knowledge cutoff. Terra: June 1, 2026. Sol: April 30, 2026. What this means: Terra has a month more built-in knowledge.
- Speed options. Terra: Fast mode at 2x. Sol: Fast mode at 2x and Ultrafast at 6x Standard in the Responses API. What this means: only Sol offers Ultrafast.
- ChatGPT access. Terra: Work and Codex on Plus, Pro, Enterprise and Edu. Sol: Work and Codex on paid plans, with OpenAI estimating 15 to 160 local messages per five hours on Plus. What this means: neither appears in regular chats.
Which fits your real work?
Entering 500 supplier invoices a week into accounting software. Terra fits: OpenAI names this exact workload, and half the token price adds up at volume.
Refactoring a payment module across a dozen files with tests. Sol fits, following OpenAI's own advice for complex coding.
A support macro that classifies tickets with no reasoning needed. Terra at none effort fits, because Sol has no option to switch reasoning off.
Cost of getting the work done
API: a job with 2 million input and 200,000 output tokens costs $3 on Terra and $6 on Sol before caching; Batch and Flex halve both. Artificial Analysis measured $0.96 per Intelligence Index task for Terra at max effort and $1.27 for Sol at max, so the gap per finished task is smaller than the token price suggests.
ChatGPT: both come with the plan's shared Work and Codex allowance, and OpenAI says the same allowance covers substantially more tasks with Terra than with Sol.
Try both on the same small task
Suggested check: run one realistic screen or document job and one coding job on both models at medium effort and compare correct results per dollar.
- Task A: every field matches your own entries.
- Task B: the test passes and no unrelated file changes.
- Record tokens for each task and multiply by each model's price.
- Repeat the pair at least twice before you settle on a model.
Decision checklist
Pick the lines that match your workload. Specs are in our GPT-6.1 Terra profile and our GPT-6.1 Sol profile.
- If most of your work is forms, invoices or drafting, start with Terra.
- If you mostly do multi-file coding, start with Sol.
- If some calls need no reasoning, Terra's none setting saves tokens.
- If you need Ultrafast latency in the API, only Sol has it.
- If you are on ChatGPT Business, check which of the two your workspace shows.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further