OpenAI releases GPT-6.1 Terra with 50% cheaper software access

One-Line Summary

AI providers are lowering the cost of routine digital work while raising vast sums to finance the infrastructure behind it.

Big Tech

OpenAI releases a cheaper Terra for everyday computer work

GPT-6.1 Terra is OpenAI's new model for reading screens and performing computer tasks, including filling forms, drafting proposals and editing videos. OpenAI makes it available in ChatGPT Work and Codex to Plus, Pro, Enterprise and Edu users, as well as through its software integration service. ¹

Standard prices through the application programming interface, or API, are50% lower than for GPT-5.6 Terra and GPT-6.1 Sol. Developers pay $1 per million input tokens and $5 per million output tokens, with tokens representing the units of content a model processes. For people using ChatGPT Work or Codex, OpenAI says the same usage allowance covers more tasks than with GPT-5.6 Terra. ¹

OpenAI reports a 15-percentage-point improvement over GPT-5.6 Terra on its professional computer-task test, at about one-third the cost at medium reasoning effort. On a separate internal test designed to provoke factual errors, the error rate at low reasoning effort falls from 53.6% to 16.7%. These are company-reported evaluations, not evidence that every office task improves by the same amount. ¹

Task-specific testing offers a more cautious picture. Yahoo Finance carries a paid DualEntry press release reporting, as of October 26, 2026, an overall accounting-test score of78.2%, but53.8% on accounts payable. The practical takeaway is to test a non-confidential example of your actual work and inspect the result before letting a faster model update business records. ²

Meta plans a $95 billion share sale for AI infrastructure

Meta plans to sell shares to finance the computing infrastructure behind its AI services, CNBC reports. The proposed fundraising totals**$95 billion**: an initial $35 billion public offering followed by a $60 billion program that sells shares into the market over time. CNBC reports that the investment supports Meta's apps, its Muse AI agent and its data-center business. ³

The New York Times frames the announcement as competition for investor funding, reporting that SpaceX also closes a $40 billion debt round for data-center chips. For workplace buyers, the distinction matters: these announcements concern financing the infrastructure behind AI, rather than a documented change to app features or employee subscription prices. Cheaper model access and expensive infrastructure investment can therefore appear in the same day's news without describing the same bill. ⁴

OpenAI expands access to fast decision-making software

OpenAI's Decisions API helps software choose quickly among supplied options rather than produce a long written answer. A user-authored October 26 post on OpenAI's developer forum discusses comparing it withJev, a competing decision model, for financial tasks. Separately, TechCrunch reports on October 22 that access opens to all customers, with voice and image support added to the public version. ^{5 6}

TechCrunch reports that language-learning service Speak uses the tool to reduce the time between actions from1.4 seconds to 0.13 seconds. For customer-support or operations teams, this is a reason to request information about a narrowly defined sorting task—not to assume the tool can safely make consequential financial decisions. Ask how incorrect choices, confidence ratings and pricing are evaluated before considering use with company data. ⁶

Industry & Biz

Fanatics adds AI outfit recommendations to product pages

Fanatics, a sports-merchandise retailer, launches an AI-poweredShop the Look feature that suggests matching, in-stock items beside a product. OpenAI's case study says Codex helps engineers build the recommendation tools, while product data, team information and Fanatics' merchandising rules guide the selections. Recommendations refresh daily as the catalog changes, and the feature is live across many Fanatics-network stores. ⁷

OpenAI reports testing recommendations across roughly 30,000 MLB products and achieving10% higher engagement than the previous experience. CNBC identifies the improvement as views of recommended items—not completed purchases. For marketers and merchandisers, the useful lesson is that relevant product data and explicit matching rules deserve as much attention as the AI-generated recommendation itself. ^{7 8}

GM targets a shorter vehicle-design cycle with AI

GM is using AI and virtual testing to change how it designs and develops vehicles, CNBC reports. The company targets roughlytwo years from concept to production, compared with a traditionalfour- to five-year cycle. CNBC notes that GM provides no timeline for achieving that goal, so the target should not be read as an accomplished productivity gain. ⁹

One concrete example is GM's reported virtual testing of more than 300,000 battery designs before building a physical prototype. For product managers and planners outside automotive manufacturing, the relevant principle is to compare alternatives before committing to expensive execution. The distinction to retain in project updates is between alternatives tested, time actually saved and a target that remains unproven. ⁹

Community Pulse

Hacker News (419↑) — Commenters find model selection increasingly complicated, while arguing that cheaper models can suffice for clearly defined, verifiable tasks. ¹⁰

"What makes this even more confusing is reasoning effort. You have Luna, Terra, Sol and Astra, each with 4+ reasoning levels. That's ~20 options to choose from. On a repetitive task like customer support or something, where you have decent benchmarks, you can plot cost versus performance, but you can't really do that on coding." — Hacker News ¹⁰

"You can do real world development with weaker models than Deepseek Flash. I still use SOTA for the main model though, but also I'm perfectly aware it's mostly out of FOMO. Deepseek is more than enough for default subagents though. Any well defined and verifiable task and there's absolutely zero difference between SOTA and good-enough models. Used to be just speed, but Deepseek is fast." — Hacker News ¹⁰

What This Means for You

A lower model price is useful only if the output is reliable enough for the task. Terra's official results and the accounting test measure different things, so compare tools using work you can check against a known answer rather than choosing solely by a general score. ^{1 2}

The Fanatics and GM examples put attention on the work surrounding AI: organizing product information, writing business rules and defining which alternatives deserve testing. For marketers, planners and designers, a useful contribution is a clear task specification and an acceptance checklist—not necessarily learning to write software. ^{7 9}

Completion also needs verification: OpenAI's Agents API documentation warns that a completed turn does not mean every tool succeeds. Before connecting workplace systems, ask your administrator which data connections and actions are permitted; OpenAI's Enterprise setup guidance explicitly calls for access decisions and security controls before broad onboarding. ^{11 12}

Action Items

1. Try Terra on a checkable draft: If you have eligible ChatGPT Work access, use GPT-6.1 Terra to revise a non-confidential proposal. Compare it with your existing draft and mark factual changes, omissions and edits you would reject.
2. Create a task-specific scorecard: Choose a routine task with a known correct answer, such as extracting details from a sample invoice. Record correctness, completion time and the manual corrections needed when testing your available AI tools.
3. Inspect Shop the Look: Open a jersey product page on Fanatics and look for matching recommendations. Check whether the suggestions respect the team, style and available stock, then write down rules that would matter for your own catalog or content.
4. Confirm workplace permissions: Ask your ChatGPT workspace administrator which data you may upload and which connected-app actions require approval. Turn the response into a short checklist before using AI with work accounts.