Google Gemini 3 Took Over The Internet

All week, we kept saying to ourselves, “Google, I wasn’t familiar with your game”—because truly, we aren’t—and the new drops from Google left our jaws dropped. If you were anywhere on Twitter this week, you must have seen a thing or two, or even a couple. If you haven’t, no worries, we got you!

Google released Gemini 3 on Tuesday. The new model scored 37.4 on the Humanity’s Last Exam, a tough test of reasoning and expertise, beating the previous record of 31.64 set by OpenAI’s GPT-5 Pro. It also took the top spot on LMArena, which ranks AI models based on how much users actually like them.

“With Gemini 3, we’re seeing this massive jump in reasoning,” said Tulsee Doshi, Google’s head of product for Gemini. “It’s responding with a level of depth and nuance that we haven’t seen before.”

The release highlights just how fast AI is moving right now. Gemini 3 comes just seven months after Gemini 2.5, less than a week after OpenAI shipped GPT 5.1, and two months after Anthropic dropped Sonnet 4.5. Google also has a more advanced version called Gemini 3 Deepthink coming soon for premium subscribers once it clears additional safety checks.

Google also launched a new coding tool called Antigravity alongside the model. It’s designed to let developers work with AI across their code editor, terminal, and browser all at once, similar to tools like Cursor. “The agent can work with your editor, across your terminal, across your browser to make sure that it helps you build that application in the best way possible,” said DeepMind CTO Koray Kavukcuoglu.

The numbers tell a bigger story too; Google says the Gemini app now has over 650 million monthly users, and 13 million developers have built the model into their workflows.

OpenAI is rolled out group chats to all ChatGPT users on Thursday, turning the chatbot into something closer to a group messaging app.

The feature lets up to 20 people join a shared conversation with ChatGPT. OpenAI says it’s designed for things like planning trips together, co-writing documents, settling debates, or working through research as a group, with the AI helping to search, summarize, and compare options along the way.

To start a group chat, you tap the people icon and invite others directly or share a link. Everyone sets up a quick profile with their name, username, and photo. Your personal settings and memory stay private, so the AI won’t spill your secrets to the group.

OpenAI says ChatGPT knows when to jump in and when to stay quiet. You can tag it directly if you want a response, and it can even react to messages with emojis and reference people’s profile photos. The launch comes just a week after OpenAI tested the feature in Japan, New Zealand, South Korea, and Taiwan. It’s available now to users on Free, Go, Plus, and Pro plans.

This is part of OpenAI’s bigger push to make ChatGPT more social. The company launched Sora back in September, a TikTok-style app where users can generate and share AI videos. Now with group chats, OpenAI is betting that AI works better as a shared experience, not just a solo one.

Google just dropped Nano Banana Pro, a big upgrade to its image generation model that can create higher-resolution images, render text more accurately, and even search the web while it works.

The new model is built on Gemini 3, which Google released earlier this week. It can generate images at 2K or 4K resolution, a major jump from the original Nano Banana, which topped out at 1024 x 1024 pixels. Google says the model is aimed at professionals who want more control over things like camera angles, lighting, depth of field, and color grading.

One of the more interesting additions is web search. You can ask Nano Banana Pro to look up a recipe and generate flash cards, for example, blending real-time information with image creation.

The model can blend up to 14 objects in a single image, use six high-fidelity reference shots, and keep up to five people looking consistent across different images. Google has released a demo app where you can try it out.

Of course, better quality comes at a higher price. The original model costs about 4 cents per image. Nano Banana Pro runs 14 cents for 2K images and 24 cents for 4K.

The model is rolling out across Google’s ecosystem. It’s now the default in the Gemini app, though free users get a limited number of generations before they’re bumped back to the original model. Paid subscribers get higher limits and can also use it in Notebook LM, Google Slides, Vids, and Google’s video tool Flow. Developers can access it through the Gemini API and Google’s new coding tool Antigravity.

Google is also adding SynthID to the Gemini app, which lets users upload an image to check if it was created or edited by Google’s AI models.

Function Health, a startup that offers regular lab testing to help people track their health, just raised $298 million in a Series B round at a $2.5 billion valuation. The round was led by Redpoint Ventures, with backing from a16z, Battery Ventures, Roku founder Anthony Wood, and NBA players including Blake Griffin.

The company’s pitch is simple: people generate tons of health data, from blood tests to wearables to doctor’s notes, but most of it just sits there. Function wants to pull it all together and make it actually useful.

Alongside the funding, Function launched what it’s calling Medical Intelligence Lab, an effort to build a generative AI model trained by doctors that can give personalized health insights based on your data. The company is offering an AI chatbot that can answer questions using your previous lab results, scans, and medical notes.

“It is not good enough to be in a world where AI exists and not be applying it to your health,” said CEO Jonathan Swerdlin. “You should be able to manage your biology.”

Swerdlin emphasized that the platform is HIPAA-compliant, fully encrypts user data, and never sells personal information. Function differentiates itself from competitors like Superpower, Neko Health, and InsideTracker by taking a device-agnostic approach that integrates lab testing, diagnostics, and clinical insights, rather than just acting as a wellness app or AI coach.

The company currently has 75 locations across the U.S. and plans to nearly triple that to 200 by year’s end. Since 2023, Function says it has completed more than 50 million lab tests. The new round brings its total funding to $350 million.

Photoshop AI editing that looks completely real. Pablo is a Photoshop plugin that adds Nano Banana (Pro) and Seedream 4.0 to your workflow.