AI news · OpenAI
GPT-6.1 Terra: what’s new, API price and Terra vs Sol
GPT-6.1 Terra arrived on October 26. See its API price, image-input capabilities, Terra vs Sol trade-offs and a document exercise with a checkable answer.
OpenAI added GPT-6.1 Terra on October 26, 2026. It positions the model around efficient visual reasoning, computer use and writing. For everyday users, the useful question is whether it can turn source material into a correct, usable answer with less editing.
At a glance
- Context window
- 1,050,000 tokens
Sources: OpenAI: GPT-6.1 Terra
- Access
- gpt-6.1-terra
Sources: OpenAI: GPT-6.1 Terra
What changed in GPT-6.1 Terra?
The model accepts text and images and produces text. Its published context limit is 1,050,000 tokens, with up to 128,000 output tokens. Image understanding is not native image generation; image tools and computer-use workflows depend on the application. Do not assume a chat app exposes every API capability.
Sources: OpenAI API changelog · OpenAI: GPT-6.1 Terra
GPT-6.1 Terra price: what does it cost?
Standard API rates are $1 per million input tokens and $5 per million output tokens; cached input is $0.05. For 10,000 uncached input tokens and 1,000 output tokens, that is $0.015 before tools or additional reasoning output. Requests over 272,000 input tokens have higher rates. API billing is separate from consumer subscriptions.
Sources: OpenAI: GPT-6.1 Terra
Terra vs Sol: choose by accepted results
Sol’s standard input/output rates are $2/$10 per million tokens. Terra therefore has half those token rates, but not necessarily half the cost per completed task. Keep the source, prompt and available tools identical. Record errors, retries, elapsed time and editing effort. Include Sol or Astra when a hard reasoning task needs a stronger comparison; a larger version number alone is not a ranking.
Sources: OpenAI: GPT-6.1 Sol
Try a document extraction task you can verify
Use this fictional invoice before sharing any real document. A correct table has A17, 40 units, 12 November and payment status unknown. The expected total is 40, not 80: an invoice amount is not proof of payment. If the model guesses that it is paid, the task fails. This is our answer key, not a recorded Terra response. A screenshot can test visual reading separately when image input is available.
Use only this fictional invoice: Invoice A17; design review; amount 40 units; due 12 November; payment status not stated. Return a table with invoice ID, amount, due date and payment status. Write unknown for missing information. Then explain what cannot be concluded. Do not infer payment from the amount or due date.Questions, answered
Does Terra generate images directly?
Its model specification lists image input and text output. A separate image-generation tool may be available in an application; that is different from the base model producing images.
Is this a Terra benchmark?
No. The example is an original exercise with an answer key. It lets you compare results without presenting invented model scores.
Can I try this exact model in GO AI?
Availability of this exact model in GO AI is not confirmed. The links open GO AI with its currently offered models, not a guaranteed selection of the model discussed here.
Sources and editorial note
Reported facts are linked to their sources. Sections marked GO AI analysis contain our interpretation; prompts are suggestions, not recorded test results.