🚀 Google Launches Gemini 3.5 Flash

PLUS: OpenAI Solves 80-Year Math Problem

Author

The Summary AI
May 22, 2026

In partnership with

Welcome back!

At its annual I/O event, Google unveiled a wave of major launches, including Gemini 3.5 Flash, now the default model across Google Search and Gemini, Omni, a new multimodal video model that accepts text, images, audio, and video, and Antigravity 2, a rebuilt multi-agent coding system. Let’s unpack…

Today’s Summary:

  • 🚀 Google launches Gemini 3.5 Flash

  • 🎥 Gemini Omni video to video

  • 🧮 OpenAI model solves famous math problem

  • 🤝 OpenAI cofounder Andrej Karpathy joins Anthropic

  • ⚡ Qwen3.7 runs in Claude Code

  • đź’° Cursor Composer cuts coding costs

  • 🛠️ 2 new tools

TOP STORY

Google launches Gemini 3.5 Flash

The Summary: Google released Gemini 3.5 Flash, making it the default model for Google Search, the Gemini app, and developer platforms. The model delivers a fast 280 tokens per second and leads in multimodal benchmarks with 84% on MMMU-Pro. Google built it specifically for agentic workflows. The model now runs on Google Search, bringing frontier AI capabilities to billions of users.

Key details:

  • Serves at 280+ tokens/sec, 4x faster than other frontier models

  • Powers new Gemini Spark personal AI agent rolling out to trusted testers, with beta coming to Google AI Ultra subscribers

  • Estimated architecture of 250-400B total parameters with only 10-16B active, running efficiently on single TPU 8i hardware

  • Introduced new Interactions API for developers

Why it matters: Google deployed its new Flash model across every major surface at once, from free Google search to Gemini platforms. The architecture reveals something fascinating: frontier intelligence no longer requires massive active parameters. Most models in this class run at 10-16B active parameters. The speed advantage matters in agentic workflows, which require dozens of interaction turns. Google also redesigned its search box for the first time in 25 years to accommodate this, expecting users will ask longer, deeper questions.

FROM OUR PARTNERS

How People Are Monetizing AI Tools

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

  • A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

  • Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

  • Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

GOOGLE

Google launches Gemini Omni for multimodal video creation

The Summary: Google released Gemini Omni, a multimodal model that generates video from any combination of inputs: text, images, audio, or video. Unlike Veo, which handles only text-to-video, Omni processes multiple inputs simultaneously to create scenes that demonstrate understanding of physics and context. Users can edit videos iteratively through conversation, swap characters with reference images, and apply motion from one video to another.

Key details:

  • Can reference multiple inputs simultaneously, combining image style references, video motion, and audio

  • Avatar creation requires a verification recording where users speak a series of numbers, with avatars stored for future use

  • API access coming in weeks, followed by Omni Pro version

  • Rolls out to the Gemini app, YouTube Shorts, Flow AI+, Pro and Ultra

Why it matters: Most video models still behave like slot machines. You write a prompt, reroll, and hope for a usable clip. Omni moves video generation closer to software editing, where scenes become editable state. Google keeps steering AI toward “world models” that simulate reality instead of predicting text.

FROM OUR PARTNERS

Effortless Voice Dictation

Your best prompts are the ones you'd never bother typing.

The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.

OPENAI

OpenAI model solves 80-year math problem

The Summary: An internal OpenAI reasoning model solved one of the most famous open problems in discrete geometry, the Erdős unit distance problem. For decades, mathematicians believed they had found the best possible solutions, but the model found an infinite family of counterexamples that beat those limits. The proof drew on ideas that experts did not expect to matter.

Key details:

  • ErdĹ‘s posed the problem in 1946 and offered $500 for a proof; the previous result remained unchanged for nearly 80 years

  • The model applied techniques in algebraic number theory traditionally considered unrelated to planar geometry questions

  • Fields Medalist Tim Gowers says he would have recommended the paper for publication in Annals of Mathematics without hesitation

Why it matters: The result worked partly because AI explored paths that many researchers ignored, combining encyclopedic knowledge with relentless willingness to test unlikely ideas. The mixing of distant theories raises questions about what other hidden links remain dormant in mathematics and science, where human attention is often limited by specialization.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/