Gemini 3.1 Pro is Google’s new “do the hard stuff” model: better reasoning, longer context, and built for workflows where you need structured outputs, not cute answers. On the ARC‑AGI‑2 benchmark – which probes for novel reasoning patterns instead of memorized test questions – it more than doubles Gemini 3 Pro’s score, which is a big deal if your app does planning, data wrangling, or agent pipelines.​

Right now you can:

Who it’s for

Highlights for our crowd

Google is explicitly pitching Gemini 3.1 Pro as the model for intricate workflows – think multi‑step planning, organizing scattered data, and supporting creative projects end‑to‑end, not just spitting out paragraphs. The ARC‑AGI‑2 jump (more than 2× Gemini 3 Pro’s score) suggests real gains in novel reasoning, which is exactly what you need for agents and automations that won’t fall apart off‑script.​

Why this matters:
If you’re building client automations, research copilots, or data‑heavy internal tools, this is Google saying “use this one when you care about not hallucinating your roadmap.” It’s a strong signal that the competitive edge is shifting from “which model sounds smartest” to “which stack actually handles weird real‑world edge cases.”

The Gemini app now ships with Lyria 3, Google DeepMind’s latest generative music model, in beta. You can type something like “afrobeat track for a reel about my food truck launch” or upload a photo/video, and it generates a 30‑second track, including lyrics if you want them, plus custom cover art via Nano Banana.​

Every track is watermarked with SynthID and you can use Gemini’s verification tools to check if audio, images, or video were generated by Google AI. For YouTube creators, Lyria 3 is also powering Dream Track, upgrading Shorts soundtracks and letting you customize the vibe while staying within platform rules.​

Why it matters:

Chrome now features Gemini in Chrome, an “always‑there AI” layer across desktop and mobile that can summarize pages, answer questions about what you’re looking at, and work with your open tabs. If you’re on Google AI Pro or Ultra, you also get auto browse, which lets Gemini perform tasks like booking a tour on a site while you watch.​

It ties into other Google products: creating events in Calendar, summarizing YouTube videos, and syncing across devices. You still control what it sees – you can pause it, limit access, delete history, and lean on Google’s security stack to protect your data.​

Why it matters:

If you’re thinking, “OK, Gemini can browse – how do I train my own agents to do the same?” you’ll like WebWorld, a large‑scale world model for training web agents, released as a research project with open resources. It simulates over a million real‑world web interactions so agents can practice tasks like booking flights, managing accounts, or navigating complex UIs in a safe environment.​

Agents trained in WebWorld rival top models like GPT‑4 on web‑navigation benchmarks, which is wild when you think about dropping a smaller, cheaper model into your own internal tools. While the full stack is research‑grade, the code and papers give devs a blueprint for building more robust “do it in the browser for me” agents.​

Check it out:

Practical angle: If your business runs on the browser (Shopify, Notion, Airtable, HubSpot), this is the kind of foundation that will power next‑gen “office agents” you can customize for your own workflows.

Confusing, jargon-packed, and time-consuming. Or quick, direct, and actually enjoyable.

Easy choice.

There’s a reason over 4 million professionals read Morning Brew instead of traditional business media. The facts hit harder, it’s built to be skimmed, and for once, business news is something you actually look forward to reading.

Try Morning Brew’s newsletter for free and realize just how good business news can be.

Check it out