AI news for Thursday, October 29, 2026: GitHub's Copilot Fusion picks your model, Google gates Gemini 4 Carbon

The best new tool picks your models for you, the best new model is locked to defenders, and Google's capital spending hit $51.4 billion in one quarter.

Agents Products3 min
GitHub Copilot Fusion now picks your model for you
GitHub made Copilot Fusion generally available on Oct. 28 on every Copilot plan: you pick an effort level, and it routes each prompt across models. The model picker just became a setting most of you can stop touching.
Why it matters If you code with Copilot, the question stops being which model and becomes how much effort. Paid plans get 30% off model costs inside Fusion for a limited time, so run it on a week of real work now and compare the bill while the discount lasts.
SRCGitHub
Models Safety3 min
Google's Gemini 4 Carbon goes to cyber defenders first
Google announced Gemini 4 Carbon on Oct. 28, calling it its strongest model, and is giving it first to a subset of cyber defenders, at intro API prices of $2 and $10 per million tokens. You can't build on it yet. The gate is the news.
Why it matters The top of the leaderboard is now released defenders first, and Google gives no date for the API. Keep building on Gemini 4 Argon, and budget for Carbon at its post-intro price of $4 input and $20 output, not the launch price.
SRCGoogle
Business Hardware2 min
Alphabet spends $51.4B in a quarter, burns $9.1B cash
Alphabet's capital spending more than doubled to $51.4 billion in the third quarter, leaving free cash flow at negative $9.1 billion, even as Google Cloud revenue rose 103%. The AI buildout no longer pays for itself quarter by quarter.
Why it matters The cheap tokens you build on are subsidized by spending like this, and now Google is running cash negative to fund it. If an AI provider's price is central to your margins, lock in rates where you can and keep a second provider ready.
SRCAlphabet (SEC Form 8-K, Exhibit 99.1)
Policy Safety3 min
New York bars AI chatbots from acting human with kids
Gov. Kathy Hochul signed a law on Oct. 28 that bans chatbot features like posing as a person or a romantic partner for minors, with fines up to $25,000 per violation. Any chatbot with those features will have to check every user's age first.
Why it matters If your product talks to the public in a human voice, New York just made age verification your problem, for adults too. The law can also force deletion of AI models trained on unlawfully collected data, so audit your companion features and your data before it takes effect.
SRCGovernor Kathy Hochul
Hardware Products2 min
Apple's touchscreen MacBook Pro runs big LLMs on device
Apple's new MacBook Pro with M5 Pro or M5 Max adds a touch OLED screen and up to 128GB of memory, from $2,899, shipping Nov. 6. Apple says M5 Max runs models with hundreds of billions of parameters locally but gives no speeds, so wait for real numbers.
Why it matters For local AI, memory decides what you can run, and 128GB fits a large quantized model on a laptop. Apple published no tokens-per-second figures, so if you are buying for local models, wait for independent tests before spending $4,499 or more on M5 Max.
SRCApple

ShareX LinkedIn Hacker News Email

Prompt of the day

Sort your AI tasks by effort so you stop overpaying

› I use AI for work and I want to stop sending every task to the most expensive model. Here are the tasks I ran through AI this past week, one per line: [list 10 to 20 tasks, with a few words on each].

Before you start, ask me one question if any task is too vague to sort.

Then, for each task:
1. Put it in one effort level: Standard (everyday work, a fast model is fine), High (needs careful reasoning), or Max (hard, many steps, mistakes are costly).
2. Pick a workflow: Single (one model answers), Cascade (a cheaper model drafts, I check it and move up a level only if it fails), or Critique (one model drafts, a second model reviews it, the first revises).
3. Write one sentence on what a bad answer would look like for that task, so I know when to escalate.

Finish with a table of every task with its level, workflow and warning sign. Under the table, name the three tasks I am most likely overpaying for today and the one task where I am probably not spending enough.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

GitHub Copilot Fusion now picks your model for you

Google's Gemini 4 Carbon goes to cyber defenders first

Alphabet spends $51.4B in a quarter, burns $9.1B cash

New York bars AI chatbots from acting human with kids

Apple's touchscreen MacBook Pro runs big LLMs on device