The Half-Price Model Goes After the Work in Front of the Computer First
OpenAI released GPT-6.1 Terra on October 26, a mid-tier model focused on visual reasoning and computer operation, with standard API pricing of $1 per million input tokens and $5 per million output tokens — half the price, OpenAI says, of the previous Terra and of GPT-6.1 Sol.
- 01THE LAUNCH — On October 26, OpenAI released GPT-6.1 Terra, positioned in the mid tier.
- 02THE PRICE — API pricing of $1 input / $5 output, cached input at $0.05.
- 03THE CHANNELS — Aimed at paid ChatGPT Work and Codex plans and the API, with same-day launches on Bedrock and Copilot.
WHAT HAPPENEDWhat Happened
GPT-6.1 Terra sits in the middle tier of OpenAI's GPT-6 family: not the flagship position, but aimed at the highest-volume repetitive work in the office — reading invoices, filling forms, reading technical drawings, drafting proposals. The model views the screen, clicks the mouse and operates across applications, turning work that once required a person sitting at a computer, cell by cell, into a single sentence in a conversation.
Price is the real headline of this launch: standard API pricing of $1 per million input tokens and $5 per million output tokens, with cached input at $0.05. OpenAI says this is half the price of the previous Terra and of the same-generation Sol. The significance of cheap is not the bill total but the deployment threshold: tasks that could only be sampled before, because each run cost too much, can now run in full.
THE TASK COSTHalf the Price per Token, Cheaper Still per Task
Per-token pricing masks the real cost: how many tokens a task consumes depends on the probability the model gets it right first time. OpenAI says Terra scores 15 percentage points higher than its predecessor on Agents' Last Exam, level with the flagship Astra, while completing tasks more than three times faster.
Third-party analysis is more direct: Artificial Analysis puts Terra's intelligence index six points above its predecessor, with cost per task down from $1.40 to $0.40. The price list is the sticker; cost per task is the receipt.
THE ERROR RATEThe Error Rate at Low Effort Is the Number to Watch
On OpenAI's internal error-prone prompt set, the factual error rate at low reasoning effort fell from 53.6% to 16.7%, and to 4.3% at high effort. That figure is closer to daily use than benchmark scores: low effort is the default tier for batch tasks.
The error rate has not fallen to zero, which means batch deployment still needs a layer of human review. Cheap makes full-volume runs possible; review decides whether full-volume runs are responsible.
THE DISTRIBUTIONLaunching on Three Platforms the Same Day Is the Distribution Signal
On launch day, Terra appeared simultaneously on the OpenAI API, ChatGPT Work and Codex, GitHub Copilot, and Amazon Bedrock. A model launch is no longer just a blog post; it is a same-day rollout into the entrances developers already use.
For buyers, the platform they already have comes before the model choice: billing, auditing and permissions systems live in the platform, and the model is just a new name inside it.
THE SHIFTWhat Changed
SIGNAL JUDGMENT — ANALYSISA Cheap Model Changes the Arithmetic of Deployment
Terra's launch moves the contest from the parameter table to the cost table: whether a model is strong enough is no longer the only question; what each task costs, and what each error costs, decides whether an agent can go live at full volume. Tokenta Signal's judgment is unchanged: work out the cost per task first, then talk about model selection.
WHAT REMAINS UNCERTAINWhat Remains Uncertain
EVIDENCEEvidence
The original report confirms the launch points: released October 26, API pricing of $1 input / $5 output, cached input at $0.05, aimed at paid ChatGPT Work and Codex plans and the API.
PRIMARY SOURCESPrimary Sources
- PYMNTS: OpenAI Unveils Half-Price AI Model to Target B2B Automation
https://www.pymnts.com/news/artificial-intelligence/2026/openai-unveils-half-price-ai-model-to-target-enterprise-automation/ - Skimwire: OpenAI GPT-6.1 Terra: computer use, pricing, benchmarks
https://www.skimwire.com/openai-gpt-6-1-terra-computer-use-model/
RELATEDRelated Intelligence
- Microsoft Winds Down Its Oracle AI Cloud Rental as Older Nvidia Chips Fall Behind
- An Unreleased Anthropic Model Proved an 18-Year-Old Quantum Conjecture, Machine-Checked in Lean
- IBM Unveils Quantum System Two in Chicago, Its Most Advanced Computer Yet, Powered by Nighthawk
- Vistra Posts $1.14B Q3 Profit as Power EBITDA Jumps Over 50%, Lifts 2026 Outlook