What's New in GPT-6.1 Terra? Half the Price and Three Times Fewer Errors Than the Previous Version

On 26 October 2026 OpenAI launched GPT-6.1 Terra. Its API price is $1 per million input tokens and $5 per million output tokens, or SAR 3.75 and SAR 18.75, half the price of GPT-6.1 Sol and half the price of GPT-5.6 Terra, which it replaces. The practical takeaway: if your company wants to automate repetitive work such as reading incoming invoices and entering them into your accounting system, or filling in forms, Terra is the new middle option; Luna is ten times cheaper for very simple tasks, and Sol and Astra remain the choice for complex coding. In a rough calculation for 10,000 invoices a month, the bill drops from about SAR 412 with Sol to about SAR 206 with Terra. This article extends our earlier comparison of GPT-6 Astra, Sol and Luna with the new tier.

What OpenAI announced, in numbers

• Price: $1 per million input tokens, $0.05 for cached input, and $5 per million output tokens. Batch processing costs half of that.
• Price comparison: GPT-6.1 Sol costs $2 and $10, GPT-5.6 Terra $2 and $12, and GPT-6 Luna $0.10 and $0.50.
• Errors: on OpenAI's internal test of prompts designed to trip the model up, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%, and to 4.3% at maximum effort.
• Coding: it solves 50.3% of Terminal-Bench 4.0 tasks at maximum effort, against 21.5% for the previous version, and OpenAI states plainly that Sol and Astra remain the better choice for the hardest coding tasks.
• Computer use: on Agents' Last Exam it scores 15 percentage points above GPT-5.6 Terra at about a third of the cost, and OpenAI says it completes tasks more than three times faster than Astra and Sol.
• Specs: a 1,050,000-token context window, 128,000 maximum output tokens, text and image input, and a knowledge cutoff of 1 June 2026.

These figures come from OpenAI's own tests. Read them the way you read a product sheet: a pointer to what to test first, not a final verdict.

Sol vs Terra vs Luna: the difference in one line each

• Luna: the cheapest, for focused high-volume tasks such as classification, summarising short messages and pulling one or two fields from clean text.
• Terra: the new middle tier, built for on-screen office work: reading invoices and drawings, filling in forms, writing proposals and correspondence.
• Sol: for complex coding and professional work that needs deeper reasoning, at twice Terra's price.
• Astra: the most capable and most expensive, for tasks nothing else can handle.

What will you pay each month? An example with 10,000 invoices

Assume your company receives 10,000 invoices a month, and that each invoice plus its instructions uses about 3,000 input tokens and 500 output tokens. That is 30 million input tokens and 5 million output tokens in total:

• GPT-6.1 Terra: $55, about SAR 206, and about SAR 103 with batch processing if you do not need the result immediately.
• GPT-6.1 Sol: $110, about SAR 412.
• GPT-5.6 Terra: $120, or SAR 450.
• GPT-6 Luna: $5.50, about SAR 21.

This is an illustrative calculation at SAR 3.75 to the dollar; real token counts depend on the quality of the invoice image and the length of your instructions. The right decision comes from a test, not the table: run 100 real invoices from your archive through Luna and Terra, and compare the errors on the total, the supplier number and the date. If Luna is accurate enough, you have saved 90%; if not, Terra is the sensible floor.

Where to find it in ChatGPT

According to OpenAI's announcement, the model is available from launch day to Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra. OpenAI says the same plan allowance covers more tasks with Terra than with Sol or Astra. If your team burns through its ChatGPT Work allowance quickly, make Terra the default for everyday tasks.

Three steps before you move your system to Terra

1. Test on your own data: do not rely on published benchmarks; a sample of 100 real documents is enough to measure the error rate.
2. Review personal data: invoices and forms carry names, phone numbers and addresses, and sending them to servers outside the Kingdom falls under SDAIA's Personal Data Protection Law and its regulation on transfers outside the Kingdom. Mask what the model does not need before sending.
3. Keep a human reviewer for critical cases: route invoices above a set amount, or with low confidence, to a staff member; a 4.3% error rate at best is still not zero.

The Origami view

Half the price does not mean swapping the model everywhere. In the projects we build we split the work: Luna for sorting and classification, Terra for reading documents and filling in systems, and Sol only where the logic needs real depth. We also design systems so the model name changes in a single setting, because OpenAI has shipped three generations in four months. If you want to automate invoices or forms in your company at a cost known in advance, contact us and we will measure the cost on a sample of your documents before any commitment. If you are comparing the small models, see also our comparison Claude Haiku 5.5 or GPT-6 Luna.

Sources

• OpenAI: Introducing GPT-6.1 Terra (launch announcement)
• OpenAI: GPT-6.1 Terra model page and pricing
• OpenAI: GPT-6.1 Sol model page, GPT-6 Luna and GPT-5.6 Terra
• OpenAI: GPT-6.1 Terra system card
• SDAIA: Guide to the Personal Data Protection Law for controllers and processors