GPT-6.1 Terra Halves the Price of Computer Use, Nvidia Backs Meta's Muse With $3 Billion, and Apple Opens Its Office Apps to Outside Agents

The chatbot that only answers questions is turning into the chatbot that does the task. This week's news is about what it costs to run that kind of assistant, who pays to build it, and which apps will let it in. All three figures below are vendor-reported or based on early coverage, so treat them as claims to test, not benchmarks.

1. GPT-6.1 Terra Makes Screen Control a Mid-Tier Feature

OpenAI launched GPT-6.1 Terra on October 26. It's a mid-tier model aimed at computer use: it looks at what's on a screen, then clicks, types and moves between apps. OpenAI says it finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol, and that it matches Astra's performance at roughly a tenth of the price per token. Coverage reports API pricing around half the previous tier's, with a 1M-token context window. It's available to Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.1-terra.

Two cautions. The headline benchmark is attributed to different tests depending on the outlet, and there are no independent evaluations yet. OpenAI also says Sol and Astra remain the better choice for the hardest coding work. What's solid is the direction: computer use has moved from a premium demo to a model you'd put on routine back-office tasks, which changes the cost math for anyone who dismissed it as too slow and too expensive.

2. Nvidia's $3 Billion Bet on Meta's Muse

Nvidia said it will invest $3 billion in Meta's equity offering while the companies deepen their partnership on Muse, Meta's personal assistant. Muse is pitched as an agent that books travel, fills in forms and makes purchases, working across email, calendar and payments. Meta plans to run more Nvidia hardware, including Vera Rubin chips, and the two will work on making the models cheaper to train and serve. Reports disagree on the size of the overall raise, and the companies gave few details on terms or timeline.

For builders, the interesting part is the alignment of incentives. The company selling the chips is financing a consumer assistant that spends users' money on their behalf. Coverage also notes Muse has already hit friction, including a security flaw in its Mac app and a block from Amazon. Agents that transact will keep running into sites that don't want them, which is a product problem as much as a model problem.

3. Apple Lets Outside Agents Into Pages, Keynote and Numbers

Apple is beta-testing an AI assistant in Pages, Keynote and Numbers that drafts documents, builds slides and writes spreadsheet formulas from natural language. Core features sit behind the paid Apple Creator Studio subscription, the beta is US English only, and reports say it needs iOS 27 or later. Sources disagree on the exact rollout date and on which models power it, with one citing a mix of Apple's own models and OpenAI's.

The detail worth your attention is Model Context Protocol support on macOS 27, which lets third-party agents connect to the apps. An assistant that can drive a document editor through a standard protocol is the same pattern enterprise teams are building internally, now shipping in a mainstream productivity suite.

Why It Matters for Builders

Each story sits at a different layer. Terra lowers the price of an agent that operates software, Muse shows who funds the consumer version, and Apple's MCP support shows applications opening a front door for agents. If you ship a chatbot, the practical questions are whether your product exposes a clean tool or MCP surface, whether you can afford to let an agent loop on a screen, and what you do when a site blocks it.

Suggested visual: a three-layer diagram (model, assistant, application) with the week's news placed at each layer. Suggested code sample: a short MCP server exposing a single append_slide tool, with a note on scoping what an agent may change.

— Maya