GPT-6.1 Terra: price, availability and how it did in my English test

What GPT-6.1 Terra is

Terra is the middle tier of OpenAI's GPT-6 family. The order from cheapest to most capable is Luna, Terra, Sol and Astra. GPT-6.1 Terra replaces GPT-5.6 Terra and sits below GPT-6.1 Sol in price and, on hard coding work, in capability.

OpenAI pitches it at everyday office work: drafting proposals, reading technical drawings, filling in forms, editing video and, through the API, reading incoming invoices and keying the details into accounting software. The headline claim is computer use, meaning the model operates application screens rather than only producing text.

Do not expect a GPT-6.1 Astra next to it. OpenAI halted that model, so the top Astra is still GPT-6 Astra. Background on that decision: why GPT-6.1 Astra was halted.

Where you can use it and which plan you need

• From 26 October 2026 Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and in Codex.
• It is not in the regular Chat view. OpenAI's help centre says plainly that Terra "is not available in Chat".
• OpenAI's announcement does not list Business, Free or Go. Until OpenAI confirms them, treat those plans as not included.
• Work and Codex share one usage allowance. OpenAI says the same allowance covers more tasks with Terra than with GPT-5.6 Terra and substantially more than with GPT-6.1 Sol or GPT-6 Astra, but gives no numbers.

Plan prices on OpenAI's US pricing page (read 24 October 2026): Go $8, Plus $20, Pro $100, $200 or $500 a month. Those are pre-tax dollar prices. If you pay from the UK or elsewhere, the amount in your currency appears at checkout and includes local VAT. On 26 October the pricing table did not list Terra yet; it showed GPT-6.1 Sol as "limited" on Plus and "extended" on Pro. More on how the shared allowance runs out: ChatGPT Plus Codex limits.

API price

The model ID is gpt-6.1-terra. Standard prices per million tokens:

• Cached input costs $0.05, which is 5% of the normal input rate. Writing to the cache costs $1.25 for 30-minute retention and $2.50 for 12 hours.
• Above 272,000 input tokens the whole request is billed at 2x input and cache rates and 1.5x output, so $2 and $7.50.
• Batch and Flex are 50% cheaper: $0.50 and $2.50. Fast mode costs 2x Standard. Regional processing adds 10%.
• A worked example: 1,000 requests with 2,000 input and 500 output tokens each is 2 million input tokens ($2.00) plus 500,000 output tokens ($2.50), so $4.50. The same job costs $9.00 on GPT-6.1 Sol, $18.00 on Claude Opus 5.5 and $2.25 on Terra through Batch.

OpenAI says Standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Against GPT-5.6 Terra that holds for input ($1 vs $2); on output the cut is bigger, $5 vs $12.

My test in British English: 71/72

I tested GPT-6.1 Terra and GPT-6.1 Sol on 26 October 2026 through OpenRouter (OpenAI's Standard endpoint, no fallback), reasoning effort medium, one attempt per task. The set is the same 12 tasks written in British English that I used for Kimi K3.1 Pro on 24 October: a PCN appeal letter, idioms, a Cardiff council news summary, translation traps, VAT maths, UK date formats, NHS number validation code, a logic puzzle, JSON extraction, UK facts, about 30,000 tokens of residents' meeting minutes and proofreading. Each task scores 0-6. Maths, JSON, facts and code are checked automatically; writing is scored blind by two AI judges, GPT-6.1 Sol and Gemini 4 Argon. Answers from OpenAI models are scored by Gemini only.

• Terra dropped a single point, on the NHS number function. The code passed all hidden tests (5 of 5 valid numbers accepted, 6 of 6 invalid ones rejected), but the docstring was one sentence and there were no comments explaining the modulus 11 steps.
• Everything else, including the formal letter to Nottingham City Council and the 30,000-token needle search, scored 6/6.
• Cost: $0.050 for all 12 tasks. GPT-6.1 Sol scored the same 71 for $0.103, Claude Opus 5.5 scored 71.5 for $0.384.
• I re-scored the older answers in the same batch as Terra, so a few rivals differ by up to 2 points from my Kimi article. One attempt per task means a 1-2 point gap is effectively a tie.

In practice: for English business writing and structured extraction, Terra matched models that cost two to eight times more in this test. In a separate Polish run of the same test it scored 65/72 (after I re-checked the logic puzzle by hand), so English is clearly its stronger side.

Benchmarks: what OpenAI claims

These are vendor results, not my measurements.

• Agents' Last Exam, medium effort: according to OpenAI, 15 percentage points above GPT-5.6 Terra at about one third of the cost, and on par with GPT-6 Astra at one tenth of its price per token.
• Agents' Last Exam, max effort: according to OpenAI, over 3x more token-efficient than Claude Opus 5.5 at a quarter of its price per token, for roughly 9x lower cost per task.
• Speed: according to OpenAI, tasks finish over 3x faster than on GPT-6 Astra and GPT-6.1 Sol.
• Error-inducing prompts (internal set): according to OpenAI, the error rate at low effort fell from 53.6% (GPT-5.6 Terra) to 16.7%, and to 4.3% at max effort.
• Terminal-Bench 4.0, max effort: according to OpenAI, 50.3% vs 21.5% for GPT-5.6 Terra. OpenAI itself still recommends GPT-6.1 Sol and GPT-6 Astra for the hardest coding.

Independent check: the Artificial Analysis Intelligence Index (read 26 October 2026) puts Terra at 48.1 on max effort and 41.3 on medium. For comparison: Claude Opus 5.5 57.6, GPT-6 Astra 52.7, Gemini 4 Argon 52.6, GPT-6.1 Sol 51.8. Terra is cheaper, not smarter.

OpenAI's system card rates Terra "Critical" for cybersecurity under its Preparedness Framework, and it ships with the same safeguards as GPT-6 Astra.

Technical details

• Context window: 1,050,000 tokens; maximum output: 128,000 tokens.
• Knowledge cutoff: 1 June 2026.
• Input: text and images. Output: text. No audio or video.
• Reasoning effort: none, low, medium (default), high, xhigh, max. OpenAI suggests starting at medium.
• Endpoints: Chat Completions, Responses and Batch. Tool calling requires the Responses API. No fine-tuning.
• Tools: web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, tool search.
• Rate limits (RPM / TPM): Build 5,000 / 1M, Launch 10,000 / 4M, Grow 15,000 / 40M.

Sources

• OpenAI: Introducing GPT-6.1 Terra - read 26 October 2026
• OpenAI API: GPT-6.1 Terra model page - read 26 October 2026
• OpenAI API pricing - read 26 October 2026
• OpenAI Help: GPT-6 and other models in ChatGPT - read 26 October 2026
• GPT-6.1 Terra system card - read 26 October 2026
• Artificial Analysis: GPT-6.1 Terra - read 26 October 2026
• OpenRouter: openai/gpt-6.1-terra - read 26 October 2026
• ChatGPT pricing - read 24 and 26 October 2026