Google Throws Shade at OpenAI With Gemini 3

Rise and compile, Tech Insiders.

AI models, weather bots, and factory droids are punching in early, while DDoS storms and zero-days pound the night shift.

Ready to tour the lab where code meets chaos?

Here's what you need to know today:

Google's Gemini 3 Fires Up Search and Apps

The new model just went from lab coat to main stage, landing in Google Search before the coffee cooled.

Google pulled the wrapper off Gemini 3 yesterday, immediately wiring it into the Gemini app, AI Mode in Search, and Google Antigravity, a new agent-first integrated development environment (IDE).

The Pro-tier model leans on a one-million-token context window (trained on data through January 2025) to parse complex prompts and generate bespoke UI tools right inside search results.

Developers can tap the API at a preview price of $2 per million input tokens and $12 per million output (for prompts under 200,000 tokens). Enterprises get the brains through Vertex AI, while Android Studio and Gemini CLI now offer agentic coding perks.

Under the hood, Gemini 3 tops the LMArena leaderboard, the popular user-voted ranking, with a record-breaking score and sets highs on PhD-level reasoning tests like Humanity's Last Exam, a grueling multimodal assessment of academic mastery. A specialized Deep Think mode pushes scores even higher (93.8% on GPQA Diamond, a benchmark for expert-level scientific knowledge). Google claims the model trades sycophancy for crisp insights.

Gemini Agent, now rolling out to AI Ultra subscribers, can triage Gmail and execute multistep workflows.

Why it matters: Gemini 3 isn't just a benchmark flex; it's already in the world's most-used search engine. If the rollout sticks, Google could claw back mindshare from OpenAI while turning every search box and code editor into an agentic copilot.

Buckle up—Search just got complicated.

Share this newsletter!

Pulse Check

Would you let Gemini Agent auto-triage and answer routine Gmail messages for you?

Results from Yesterday's Pulse Check

If an app twisted a family photo into nightmare fuel, you'd...

Insider Intel

Storm AI Forecasts Get 8 Times Faster

Mother Nature just got a livestream—and it runs on TPUs.

Google DeepMind's new WeatherNext 2 model cranks out global forecasts eight times faster than its predecessor while upping accuracy on 99.9% of key variables, such as wind and humidity.

A fresh Functional Generative Network sprinkles controlled "noise" into the network's core math. This lets the system spin up hundreds of physically coherent weather scenarios in under a minute on a single TPU rather than waiting hours for a supercomputer grind.

That speed jump matters. Meteorologists can now peek 15 days ahead at hour-level resolution, spot low-probability but high-impact events, and issue earlier warnings.

The upgrade is already baked into Search, the Gemini app, Pixel Weather, and Maps' Weather API (with the standard Maps app following in the coming weeks), so your phone gets sharper storm intel by default.

Experimental versions tested this season already nailed a Category 5 intensification window for Hurricane Melissa a full day before legacy physics simulators, buying Jamaica crucial prep time.

Researchers and businesses aren't left out. WeatherNext 2 data is live in Earth Engine and BigQuery, while custom model inference is available via Google Cloud's Vertex AI.

Expect energy grids to lean hard on the model for predicting wind-farm output. However, experts warn these tools are only as good as the historical data they devour, a pipeline currently threatened by staffing shortages that one former National Hurricane Center chief warned is akin to "shuffling the deck chairs on the Titanic."

Hopefully, your picnic plans will now only be ruined by your own scheduling, not the forecast.

Flash Bytes

Security Alerts

Zero-Day Hack Hits Logitech

Logitech confirmed hackers exploited a third-party zero-day to steal internal data. An SEC filing admits the breach likely exposed employee and customer records, though the company "does not believe" credit cards or IDs were stored there.

Security pros link the raid to the notorious Russia-linked Cl0p cybercriminal gang, which claims it lifted 1.8 TB via an Oracle E-Business Suite flaw, CVE-2025-61882. Logitech patched once the vendor fix dropped and says operations are untouched.

It's a reminder that tight in-house security can crumble when a vendor's code springs a leak. Expect longer questionnaires at your next renewal.

When ransomware crews start bragging on social, it's never because your mouse has extra buttons.

Azure Swats 15.7 Tbps Botnet Barrage

Azure shrugged off a blistering 15.72 terabit-per-second distributed denial-of-service (DDoS) attack on Oct. 24, the platform's largest ever. Targeting a single Australian endpoint, Microsoft says the sudden junk-data bursts hit 3.64 billion packets per second without knocking services offline.

Investigators traced the onslaught to Aisuru, a Mirai-based botnet weaponizing over 500,000 hacked home routers and cameras. This same herd pushed a record 22.2 Tbps blast at Cloudflare in September, showing how attacks surge alongside faster consumer fiber speeds.

Microsoft says holiday traffic is prime DDoS season, so stress-test defenses now. Drill-ready playbooks beat panic patching at 3 a.m.

Reality check: If your "smart" doorbell has a busier social calendar than you do, check its firewall.

Industry Shakeups

China Deploys Walker S2 Factory Humanoids

UBTECH has begun mass-shipping hundreds of its Walker S2 humanoid robots—China's first large-scale rollout of full-size mechanical coworkers.

Standing 1.76 meters (about 5 feet 9 inches) tall, each bot sports 52 degrees of freedom (DOF), dexterous 11 DOF hands, a 10 cm (4-inch) circular face screen, and a self-swapping dual-battery pack that lets it work 24/7 without human pit stops.

Orders booked in 2025 now top 800 million yuan ($112.5 million) from automakers like BYD, Geely, Audi FAW, and FAW-Volkswagen, plus logistics giants Foxconn and SF Express.

Early deployments have robots deep-squatting to lift 15-kilogram (30-pound) parts, shuttling components across assembly lines, and detecting anomalies in smart-factory dark zones where humans rarely tread.

UBTECH plans to deliver 500 units by year-end, ramp production to 5,000 annually in 2026, and hit 10,000 by 2027. The company has shifted from selling standalone bots to turnkey scenario kits combining its BrainNet control platform, AI task planning, and on-site training, aiming to make humanoids plug-and-play for industrial customers.

The push gives China a head start in commercial humanoids while major rivals like Tesla's Optimus, Figure AI's 03, 1X's NEO, XPeng's IRON, and Unitree's H2 remain in pilot or preorder phases.

Investors have noticed: UBTECH's Hong Kong-listed shares have soared 150% this year, even as skeptics, including Figure AI's CEO, flag mismatched light reflections in publicity videos as potential CGI foul play.

Someday your factory tour might include a warning: "Mind the robots—they don't take coffee breaks, but they do steal your step count."

Meet Our Author

Justin Meyers

Writer at TechnologyAdvice

Justin Meyers is an investigative writer and editor who draws on over a decade of meticulous hands-on research to deliver the full, trustworthy story behind consumer and enterprise tech, including cybersecurity.

Curious about where AI is really headed?

The Neuron cuts through the noise to bring you smart, hype-free takes on the latest AI trends, tools, and breakthroughs. Join 500,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

Advertise in Daily Tech Insider!

Daily Tech Insider is a TechnologyAdvice business

© 2025 TechnologyAdvice, LLC. All rights reserved.

TechnologyAdvice, 3343 Perimeter Hill Dr., Suite 215, Nashville, TN 37211, USA.