Google’s Flash-y new Gemini 3 release

PLUS: Comparing ChatGPT Image and Nano Banana Pro

Good morning, AI enthusiasts. OpenAI may have been riding high into the holidays after back-to-back releases, but Google just crashed the party again with another flashy gift of its own.

The company’s new Gemini 3 variant combines top-level intelligence with extreme speed at a fraction of the price — a combo that may be the toughest yet for its AI rivals to answer.

In today’s AI rundown:

  • Google’s Flash-y new Gemini 3 release

  • Amazon discussing $10B+ investment in OpenAI

  • Comparing ChatGPT Image and Nano Banana Pro

  • Stanford AI experts predict 2026 will be a year of reckoning

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

GOOGLE

Image source: Google

The Rundown: Google just rolled out Gemini 3 Flash, a speed-optimized version of its recently released flagship model that still maintains frontier-level intelligence, becoming the new default model across Gemini and in Google Search’s AI Mode.

The details:

  • Gemini 3 Flash matches and even exceeds 3 Pro across a range of benchmarks, while coming in at ¼ the price and 3x the speed.

  • On Humanity’s Last Exam, Flash scored 33.7% — tripling the 11% from its predecessor and nearly matching GPT-5.2’s 34.5%.

  • Both the Gemini App and Google Search’s AI Mode now default to 3 Flash, combining real-time web results with fast, improved reasoning.

Why it matters: It may sound counterintuitive, but Gemini 3 Flash feels like a bigger deal than 3 Pro — offering a unmatched intelligence and speed combo at prices that significantly undercut the competition. Google continues to eat away at OpenAI’s market share, and Flash is looking like yet another reason for that trend to continue.

TOGETHER WITH ATLASSIAN ROVO

The Rundown: Discover Atlassian Rovo — AI that knows your business. Rovo connects teams, knowledge, and tools so you can move faster and work smarter — together.

Why Rovo?

  • Rovo connects to all your favorite SaaS apps.

  • Rovo brings organizational knowledge and context into every workflow.

  • It’s already built into Jira, Confluence, and more.

  • And it’s built on Atlassian’s enterprise-grade security & privacy.

AMAZON & OPENAI

Image source: Nano Banana Pro / The Rundown

The Rundown: Amazon is reportedly negotiating a potential $10B investment in OpenAI that would value the AI leader above $500B, according to The Information — with the deal also possibly including a commitment to use Amazon’s Trainium AI chips.

The details:

  • The two companies signed a 7-year, $38B AWS cloud contract last month, with OpenAI now partnered with “at least” 5 cloud providers.

  • OpenAI would adopt Amazon’s Trainium processors, giving AWS a high-profile customer for chips competing against Nvidia.

  • The companies have also discussed commerce and enterprise partnerships, with OAI continuing to position ChatGPT as a shopping destination for users.

Why it matters: The restructuring that freed OAI from Microsoft exclusivity is paying off, with the ability to court competing cloud providers for massive infra needs. For Amazon, the move would hedge its Anthropic bet — while also securing a big customer for Trainium chips that have struggled to gain traction against Nvidia’s dominance.

AI TRAINING

The Rundown: Learn the key similarities and differences between Google’s Nano Banana Pro and ChatGPT’s new image generation model, while also learning how to build your own comparison matrix that you can reuse for other model comparisons.

Step-by-step:

  1. Pick 5 use cases (ours were logo, website graphic, IG post, marketing brochure, photorealistic image) and outline testing rules: same prompt per model, 4 images each, graded 1-5 on consistency, creativity, utility, and quality

  2. Feed use cases into Claude/ChatGPT/Gemini and prompt: “Here’s my use cases: [X]. Write me a json prompt for each with 4 variations in a 4x4 grid”

  3. Create a scoring matrix (duplicate our Notion guide here) where Overall Rating = (Consistency + Creativity + Utility + Quality)/4

  4. Generate images by putting each prompt into both tools using new chats per use case, then rate outputs based on your criteria

Pro tip: To save on time/tokens, tell the LLM to write a prompt that will generate 4 variations in a 4x4 grid.

PRESENTED BY BOX

The Rundown: 60% of enterprises expect AI transformation within 2 years, and Box’s Executing AI-First Series is the step-by-step playbook for empowering teams to thrive in the era of AI.

In this series, you’ll learn:

  • How Box approached becoming AI-first through its value realization strategy

  • How to Deploy agents with an ideate>pilot>rollout>scale plan

  • How to be an AI manager

  • How to measure what matters by tracking AI agent impact

Read the first article in Box’s series and follow along for actionable insights and downloadable templates.

STANFORD UNIVERSITY

Image source: Nano Banana Pro / The Rundown

The Rundown: Stanford University’s Institute for Human-Centered Artificial Intelligence (HAI) just shared its AI predictions for 2026: a “ChatGPT Moment” for healthcare and a shift from hype to hard evaluation of what AI can actually deliver.

The details:

  • HAI Co-Director James Landay predicts “no AGI this year,” expecting more companies to admit AI hasn’t delivered gains outside coding and call centers.

  • Economist Erik Brynjolfsson forecasts a rise in “AI dashboards” tracking displacement and productivity at the task level updated monthly, not years later.

  • Researcher Curtis Langlotz anticipates a “ChatGPT moment” for healthcare as the training cost of medical models decreases and dataset accessibility rises.

  • Law professor Julian Nyarko said firms will move from “Can it write?” to “How well, on what, and at what risk?” with a shift toward more complex legal work.

Why it matters: With 2025 being the year of AI hype and massive investments, the experts see 2026 as a transition to asking whether it was worth it. Stanford’s faculty isn’t expecting an AI bubble crash like many others, but their predictions do point to an industry that may have exhausted its patience for overpromising demos and pilots.

QUICK HITS

  • 🔐 Incogni - Erase sensitive data like addresses and phone numbers from the web. Get 55% off with code RUNDOWN*

  • ⚡️ Gemini 3 Flash - Google’s powerful, cost-effective frontier reasoning model

  • 🌌 ChatGPT Images - OpenAI’s upgraded image generation system

  • 🗣️ Chatterbox Turbo - Resemble AI’s fast, expressive, open-source TTS model

*Sponsored Listing

Alibaba unveiled Wan2.6, a new multimodal model that can generate up to 15 seconds of HD video with dialogue, storyboarding, and character reference capabilities.

U.S. Sen. Bernie Sanders announced plans to pursue a pause on AI data center construction, citing concerns about job displacement and societal impacts.

Amazon’s Peter DeSantis will now lead a new division overseeing AI models, chips, and quantum computing, while Alexa and Nova architect Rohit Prasad departs.

xAI introduced the Grok Voice Agent API, allowing developers to build voice tech using the company’s top-ranking speech-to-speech model.

Meta released SAM Audio, a model that can isolate specific sounds from audio or video files using text descriptions, visual clicks, or timeline selections.

COMMUNITY

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Ritesh K. in India:

“I built an autonomous ‘Financial Firewall’ that reads and audits invoices better than a human team. Instead of using brittle OCR templates, I use Vision Models to extract data from messy PDFs, but the magic is the semantic reasoning.

The AI compares every invoice against the original Purchase Order, and understands context. Crucially, it also catches subtle exceptions that human reviewers often miss due to fatigue — like a vendor slipping in a shipping fee on a ‘free shipping’ order or small unit-price variances. It auto-flags these discrepancies with a summary of the error, ensuring zero financial leakage while fully automating the boring approval logic.”

How do you use AI? Tell us here.

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Reply

or to participate.