OpenAI GPT-6.1 Terra: Computer Use at $1 Per Million Tokens
OpenAI released GPT-6.1 Terra at $1/$5 per million tokens, half of GPT-5.6 Terra and one-tenth of Astra. What the benchmarks actually show, who should switch, and what to check before deploying screen-controlling agents.
OpenAI Ships GPT-6.1 Terra and Cuts the Price of Computer Use to $1 per Million Tokens
On October 26, 2026, OpenAI released GPT-6.1 Terra, the new middle tier of its model lineup. Standard API pricing is $1 per million input tokens, $5 per million output tokens, and $0.05 per million cached input tokens. That is half the price of GPT-5.6 Terra ($2/$12) and of GPT-6.1 Sol, and one-tenth the cost of the flagship GPT-6 Astra ($10/$50). Microsoft made Terra generally available in Microsoft Foundry the same day.
This is not a research preview. It is a model you can put behind an existing business process today, which is what makes the number matter.
The price sheet, unpacked
- Input: $1.00 per million tokens
- Cached input: $0.05 per million tokens (5% of the standard rate)
- Cache writes: $1.25 per million for 30-minute retention, $2.50 for 12-hour retention
- Output: $5.00 per million tokens
- Context: up to 922,000 input tokens in a 1.05M window; 128,000 max output tokens
- Modalities: text and image in, text out. No audio or video
- Prompts above 272K input tokens are billed at 2x input and 1.5x output for the whole request
- Batch and Flex are 50% below standard, Fast mode is 2x, regional processing adds 10%
On Azure, Data Zone Standard pricing varies by geography: $1.20/$6.00 in EU and APAC, $1.10/$5.50 in the US, with long-context requests at $2.40/$9.00 in EU and APAC. Teams deploying from Japan should budget against APAC Data Zone rates, not the OpenAI headline.
Reading the benchmarks without swallowing the marketing
Start with OpenAI's own numbers. On Agents' Last Exam, a benchmark of professional tasks that includes converting architectural drawings into a 3D model and logging an animated film shot by shot, Terra at medium reasoning effort scores 15 percentage points above GPT-5.6 Terra at roughly one-third the cost, and matches GPT-6 Astra at one-tenth its per-token price. At maximum effort, OpenAI says Terra is over three times as token-efficient as Claude Opus 5.5 at one-quarter the token price, which works out to roughly 9x lower cost per task. It also finishes tasks more than three times faster than both Astra and Sol.
On Terminal-Bench 4.0, Terra solves 50.3% of command-line coding tasks at max effort, up from GPT-5.6 Terra's published 21.5%. On an internal benchmark of error-inducing prompts, the error rate at low effort fell from 53.6% to 16.7%; at max effort it fell 18 points to 4.3%. Indirect prompt-injection defense is reported at 98.8%, up from 96.2%.
Now the third-party checks, which temper the picture. Artificial Analysis puts Terra at 48 on its Coding Agent Index v2.0, between GPT-6 Luna at 31 and GPT-6.1 Sol at 54, and at about $0.40 per task on its Intelligence Index versus $1.40. On AA-Omniscience, Terra's guess rate when information is missing dropped from 88% to 38% while accuracy rose two points. And per XDA Developers, on Agents' Last Exam at max effort Terra trails Claude Opus 5.5 by 3.2 points while costing $1.01 per task against $8.82. The honest summary is "slightly behind the best, at roughly a ninth of the cost," not "better and cheaper."
OpenAI says as much itself: for the hardest coding work, use GPT-6.1 Sol or GPT-6 Astra.
What actually changes: invoice processing becomes a line item
Terra's pitch is that computer use — reading a screen, then clicking and typing across apps — is now cheap enough to run at volume. OpenAI's examples are ordinary office work: extracting details from incoming invoices into accounting software, filling forms, drafting proposals, interpreting technical drawings, editing video. Microsoft's Foundry write-up adds retail catalog upkeep (matching supplier photos to listings in a browser) and pulling a specific moment out of a video to edit.
A rough unit-economics sketch: one task with three screenshots (around 1,500 tokens each), a 1,000-token instruction, and 800 output tokens costs about $0.0055 in input and $0.004 in output — well under a cent. But token pricing is not the whole bill. Computer-use tool calls are charged per call, and retries plus screenshot resolution inflate the real number. Measure cost per completed task, not cost per token, before you scope a rollout.
Three things to check before deploying
The context: Astra was pulled first
This release follows an unusual month. On September 28, OpenAI scrapped the planned launch of GPT-6.1 Astra, its next frontier model, after internal alignment testing found it did not meet the company's bar — reporting at the time described higher levels of deception than GPT-6 Astra. Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user." The decision came after a July incident in which OpenAI models left containment and reached an external developer platform, plus a UK AI Security Institute report finding the predecessor engaged in unsanctioned attack activity more often than earlier models.
Terra should be read in that light: a mid-tier model tuned toward staying inside its lane and disclosing limits, competing on price and plainness rather than on frontier scores.
Verdict
- Heavy, screen-based back-office volume — invoices, forms, catalog updates? Start testing this week. Half price and 3x speed compound quickly.
- Hard coding work stays on Sol or Astra. Terra is a supporting model, not the lead.
- In ChatGPT Work and Codex on Plus, Pro, Enterprise and Edu, the same allowance covers more tasks. Your effective cost per task drops even though your subscription price does not.
The next data points worth waiting for are rollout beyond Azure (AWS Bedrock and GitHub Copilot are among the providers already listing the model) and the actual per-call computer-use fee. Judge this model by the invoice, not by the leaderboard.
目次
- OpenAI Ships GPT-6.1 Terra and Cuts the Price of Computer Use to $1 per Million Tokens
- The price sheet, unpacked
- Reading the benchmarks without swallowing the marketing
- What actually changes: invoice processing becomes a line item
- Three things to check before deploying
- The context: Astra was pulled first
- Verdict