willdepue는 DeepSeek, Moonshot, MiniMax, Ant Group 등 중국 상위 AI 연구소들이 수출이 금지된 NVIDIA Blackwell 칩으로 최신 모델을 학습했다는 포렌식 증거를 가중치 초기화 패턴의 GPU 지문 분석으로 제시하고 실제 하드웨어로 검증했다고 주장한다.
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpoint-for….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
This new Reuters report is stunning.
A week before the U.S. captured Venezuela's leader, JD Vance reportedly met at a private Montana ski club with John Addison, an oil trader who had given $300,000 to help elect Vance to the Senate in 2022. Vance allegedly asked how fast Venezuela's oil could be sold if America took control of it.
Both men deny the meeting.
Flight records show their planes parked together that day.
Two days after the Venezuela operation, Energy Secretary Chris Wright called Addison and offered his company, Vitol, the job of selling Venezuela's oil. There was no public bidding. Vitol and one rival have since made hundreds of millions of dollars.
Addison and his wife also gave $7 million to help elect Trump, and he now chairs a board advising Secretary Wright.
Congress has a duty to act.
We should bring Secretary Wright, Vitol, and Addison in to testify under oath. We should demand the phone records, emails, and flight logs. We should find out whether anyone got a heads-up about a military operation. And we should require open, competitive bidding before any company profits from our foreign policy.
If nothing was wrong, let’s see the records.
Introducing Wave.
The world's first AI that reads your customers' mind.
$1.47 trillion will be spent on software this year, but only a fraction of software features are used by customers.
AI can build anything you want, but it doesn't know what your customers need.
Wave fixes that.
Disengaging from syntax is not disengaging from engineering.
I’m working on a significant project at the moment. It’s in a pre-production state so far. More than a prototype, less than ready for real users.
I’ve been making multiple passes over the structure for each of the following concerns: Architecture, coverage, race & concurrency, security, API, etc. Each of these requires significant interrogation and review of verification procedures; but not review of the code.
At base are the /ub-mode (GitHub.com/unclebob) rules that keep the agents disciplined as far as coding structure and tests are concerned. But above that it’s me making sure that all the bases are covered well enough to keep my client safe from stupid sloppy mistakes.
I am the engineer. I am responsible. The agents are just my tools. They are power tools, to be sure. But power tools in the hands of novices result in missing fingers. So I am very careful.
We got into @ycombinator to fix how AI makes music.
Today we're launching Taan v1-preview, the most controllable foundation music model.
Music models today make catchy tracks, but you can't steer them. Ask for "violin at exactly the 11th second" and they usually skip it. So you regenerate over and over and hope one take fits.
Taan is trained to fix this.
Hum a melody, name exact notes, or hand it an image, a MIDI file, a video or a script. Taan composes music that fits your context and follows your instructions.
v1-preview has genre support limited to orchestral and cinematic music, with more genres on the way.
live today 👇
Retell just crossed $100M in revenue run rate.
$10M to $100M in 17 months. Profitable. $4.6M raised.
From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us.
Our bet: the future of the contact center belongs to the teams who run it.
Proud of our team. Grateful to every customer who believed in us.
You might want to make sure you have a few open-weight models backed up and some basic gear to run them. Who knows what really happened at NVIDIA, but the world just lost one of free AI's biggest proponents.
We’re rolling out Orchestrator in @Antigravity to Google AI Ultra subscribers, letting you run multiple models in a single session instead of routing the entire task to one model.
Orchestrator pairs a frontier model with a fast execution model: Carbon sets the plan, hands off code edits and testing to Hydrogen, and steps back in to repair the plan if it hits friction — reaching near-Carbon task success rates on Terminal-Bench 4.0 while consuming 24% less quota and completing 19% faster on average.
Introducing intelligent eyewear, built in partnership with Samsung and our iconic eyewear brands, Gentle Monster and Warby Parker. Now available for pre-order. Learn more at goo.gle/glasses_x
Today, Search is getting a massive upgrade with Gemini 4 Hydrogen! It’s the biggest model improvement to Search yet. And it’s available now for all users, globally.
Why it’s awesome:
- Crazy fast
- Cutting-edge intelligence, with advanced reasoning, more precise instruction following, and better tool calling
- More visual responses, including images and custom tables, charts, and simulations
- More scannable response layout
- Better understanding of intent (including for long multi-part queries with ambiguous typos)
- Broader range of links to the web (improved reasoning -> more query fan-outs)
TL;DR: powerful and fast. Take Hydrogen for a spin in Search and let me know what you think!
Gemini 4 Hydrogen is here, bringing cutting-edge intelligence to everyone for fast, everyday help.
It makes it easier to get things done, following your instructions more closely, calling the right tools, and getting you to a reliable output with less back-and-forth. 🧵
Inkling 2 Small’s weights are now available for download on HuggingFace under an Apache 2.0 license.
Make Inkling your own!
huggingface.co/thinkingmachin…
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows.
It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
Sad but still bullish on @nvidia!
I’m seeing a lot of people criticizing its structure and the lack of a clear CEO successor. Having seen it from up close, I think a lot of this criticism is ill-informed and overly simplistic (as often happens when a company is run differently from the norm).
Obviously, important to remember that this structure made NV as successful as it is and powered the AI revolution. But beyond its past success, I think it’s also what makes the company antifragile for the future.
In my opinion, NV's distributed leadership and comfort with chaos are much more of a strength than a weakness, even without Jensen. They give an unusual number of leaders room to step up and keep the whole machine moving (as we've seen in the past few days). They also keep the company flexible enough to adapt to the next paradigms (open-source/local/agents/robots/biology/chemistry/...). They also encourage hard-working, self-motivated team members with a strong sense of belonging and a willingness to help each other which will matter a lot.
Personally, with the right CEO (and NVIDIA has plenty of exceptional leaders internally who could take on that role), I believe NV can keep thriving and empower every company to become an AI company and every builder to become an AI builder.
To the @nvidia team: sending you strength and support, from me, the @huggingface team and so many people across the AI community. We need you and we’re rooting for you to carry @JensenHuang's mission forward! Taking care of his company, your company, is probably the best way to honor his legendary legacy!
💚💚💚
Google Search is getting its biggest model update yet with Gemini 4 Hydrogen — rolling out globally for all users today.
Here’s what’s new in Search:
🔍 Better understands intent: Ask nuanced, multi-part questions in your natural flow of thought — even with specific requests or typos — and get relevant responses across all supported AI Mode languages.
👀 More visual responses: Get adaptive, easy-to-scan layouts with images, comparison tables, and custom tools.
🌐 Upgraded query fan-out: Hydrogen’s reasoning upgrades how Search runs simultaneous subtopic queries to connect you to a broader range of helpful web links.
Announcement: CONDUCTOR IS BACK
We’ve decided to continue development of Conductor!
Reports of our death have been greatly exaggerated (by us). Regularly scheduled updates will resume next week!
🫡🚂
Today, we’re rolling out Gemini 4 Hydrogen in Google Antigravity, bringing cutting-edge intelligence for everyday coding and agentic workflows. Currently available to paid plans including Pro, Ultra, and enterprise customers.
It stands strong, scoring 51 on the Artificial Analysis Intelligence Index (surpassing Argon’s 50, and reaching 54 at its highest setting) and 57% on Terminal-Bench 4.1 (more than tripling 3.8 Flash’s 18.2%).
Watch how we are using Hydrogen to maintain codebases
Meet Gemini 4 Hydrogen.
It brings cutting-edge intelligence to everyone, with speed and efficiency for everyday coding and agentic work.
Across 4 thinking levels, it stands on its own or teams up with Carbon and Argon — 2.5X faster than 3.8 Flash, outpacing competing models in its class.
Very excited to release Gemini 4 Hydrogen! Incredible speed and intelligence for everyday coding, agentic work and all-round use. I think you'll love this model. More info in the blog - enjoy!
Introducing Gemini 4 Hydrogen, an extremely capable and fast model built to bring Gemini 4 to everyone!
Available to all tiers of Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro + Ultra subscribers in Google Antigravity.
What does sci-fi get right and wrong about outer space?
At 2pm ET (1800 UTC), our Astrophysics Division Director Dr. Shawn Domagal-Goldman will answer this question and others during a @Reddit Ask Me Anything.
reddit.com/r/scifi/s/YWdW…
Tethys and Dione showed up for the group photo!
In addition to the two moons shown here, Saturn has a whole entourage of natural satellites—more than any other planet in our solar system. As of now, the ringed planet has 293 confirmed moons.
New @a16z market map alert! 🚨
Consumer agents have emerged as the hot category...and as a can't-miss distribution channel
We've seen products reach hundreds of thousands of users faster than ever before (with best-in-class retention!)
But, we're still SO early 👇
Building software takes more than code generation. Here's what an effective agentic engineering workflow looks like with GitHub Copilot 👇 x.com/i/article/2120…
Register for Get Certified: Special Edition to earn a brand new badge!
Our hybrid Professional Agentic Architect (PAA) cert includes no-cost prep & hands-on Google Skills challenge labs.
Google Cloud customer email required. Apply by Dec 9 → developers.google.com/gear/getcertif…
If you want to know where AI in law is going, look at software engineering. It’s about six months ahead.
Our engineers at @WeAreLegora are shipping 3x as much as they were this summer. Nothing about the team changed except how they work.
Here are a few examples of what that has produced over the past 6 months:
The Legora Agent: An AI agent that plans, executes, reviews, and delivers complex legal work end to end, with the lawyer in control where judgment is required.
Skills: Instructions teaching the Agent your standards, processes, and preferences, with 250+ ready-made Skills to start from.
Intake: The agentic legal front door for in-house teams, which turns ad-hoc requests into tracked work executed by the Agent.
Lists: A structured place to manage legal work, populated by the Agent, cited to source, and shared by the whole team.
Monitors: Regulatory horizon scanning across the jurisdictions and topics you choose, with continuous scanning and a single feed of what changes.
Ontological Legal Research: Legal research built on comprehensive legal data and a proprietary ontology that maps how sources connect, so the Agent sees the full picture of current law (live in the US).
Integration ecosystem: Permission-aware connectors linking the Agent to the systems legal teams already use, such as iManage, NetDocuments, SharePoint, and Box.
What we learned along the way is that the bottleneck moved. Writing code used to be the slow part. Now it’s fast, and the slow part is reviewing what the agents wrote. Our best engineers are spending less time typing and more time deciding whether the output is right.
I expect the same thing to happen in law. The first phase is manual: every task needs a person to start it, do it and check it. We’re now in the second phase, where work is faster and better but the operating model stays the same. In the third phase, AI takes a first pass at all work, or does it end to end where it’s permitted. Roles change at that point.
Review becomes a more important part of the job, and the legal teams that start building that muscle now will have a head start.
designers, consider this a sign to start animating with opus 5.5
let the agent run wild with your ideas, leave comments for it to iterate, fine-tune with keyframes and generated sliders
anyone can make motion graphics with agents. in @StewdioAI, you can make them yours
AI is expanding what game developers can afford to attempt.
The opportunities fall along two dimensions: whether AI is helping with an existing game or a new one, and whether its main contribution is lower cost or new possibilities.
Indie devs can now do more with lean teams, toolmakers can decompile and modify existing games, and studios can take riskier bets on original IP and experimental genres.
Some of gaming's biggest franchises began with someone tinkering with an existing game. AI gives more people a place to start.
Full piece from a16z's Elena Burger, Anish Acharya, and Jon Lai: a16z.news/p/gamings-next…
AI is making games faster and cheaper to build, giving developers more room to take creative risks, revive overlooked genres, and let players reshape the worlds they play in.
In this conversation, a16z's Jon Lai and Anish Acharya join Elena Burger to discuss why human taste still determines what makes a game fun, how modding created some of gaming's biggest franchises, the opportunity for new distribution platforms, and games that adapt to whatever players choose to do.
0:00 Intro
0:57 How AI can now build a game
6:21 What shorter dev cycles unlock
10:08 Can AI make a game fun?
12:27 When creating is the game
14:01 Building a Habbo Hotel record shop
16:16 His son rebuilt Ghost of Tsushima in Minecraft
17:54 Mods: where billion-dollar franchises come from
20:36 Decompiling and reviving dead games
22:45 The renaissance of niche genres
25:51 Managing AI agents like an RTS game
28:02 The business of AI games
33:53 What excites and scares studios
35:16 The never-ending game
37:56 The games they'd finally buQild
YouTube: youtu.be/HDRy6XmE2RQ@VirtualElena@illscience@Tocelot
Today, we are excited to introduce Ladon, a state-of-the-art biological foundation model that predicts disease trajectory and treatment response in patients. There is immense complexity in the individual biology of every patient. Two patients can have the same diagnosis and receive the same treatment, yet have very different outcomes: one responds well and lives for years, while the other’s disease continues to progress. Understanding and resolving this heterogeneity to predict outcomes is how we accelerate our delivery of medicines for patients and unlock clinical trial abundance.
Ladon is our first step towards this future: one where we can learn more from every patient enrolled so that we can run smaller, faster trials with higher success rates. We built it to give drug developers quantitative support on billion-dollar clinical trial decisions. Today we're excited to share Ladon's capabilities in our blog and technical report.
Ladon is a 3B parameter model trained on millions of RNA-seq samples spanning decades of technical and biological variation across tissues and species. Ladon beats gene expression baselines, established biomarkers, and existing RNA foundation models at predicting patient prognosis and therapeutic response in 140+ oncology tasks, including external validation on past clinical trials. This is significant because clinical trials are fundamentally generalization problems. We show that Ladon-derived predictions are robust across indications, treatments, and sequencing platforms.
Reach out if you’d like to get started with the model today!
Read more about Ladon in the blog: blank.bio/introducing-la…
Find more technical details in the white paper: blank.bio/introducing-la…
Introducing Summer Engine (YC F26). Summer makes it easy for everyone to make and play games together!
You ask for the game you want. Summer builds it, and you play it with your friends.
Last Christmas I tried to make a game with AI. The game looked great. But, then I wanted real multiplayer, analytics and monetization, and it became a huge pain! And don't get me started on publishing.
A year later it is still the same for most people. The games look much better, but you still get stuck with a prototype.
So we fixed it.
Today, tens of thousands of creators from all over the world have built games on Summer. The community has made more than 240,000 assets that every user can build with, making it one of the biggest free asset libraries in the world.
You own the game you make on Summer. You can sell it and publish it on Steam, Web, iOS, Android, and on Summer Games.
Making games should be as fun as playing them.
Try it at summerengine.com
PS. @garrytan made a game in a weekend about being a founder in the YC batch. You can try "FounderStar" on the link above.
Today at @KeyframeLabs we're releasing our most capable real-time avatar model yet, setting a new standard for realism, expressiveness, and scalability.
persona-2.0-live is the only model in the world with precise, word-level control over expression, enabling some truly delightful user experiences.
2.2-4x ahead of competition. Just a few cents per minute. 1000s of concurrent sessions. See for yourself ↓
The world doesn’t need another Harvey. It needs a Mike.
Our AI legal agent remembers your business, takes the first pass on contracts and helps you find vetted lawyers.
Built by @IndexLawHQ. Backed by @ycombinator.
Give it to Mike: usemike.com
The European Commission is ordering Google to share the search data of all Europeans with other companies (AI chatbot makers, search engines, etc.), after a kind of "anonymisation". I tested it on the real Google data, hundreds of billions of search records. I broke it.
Findings. I singled out the people behind half a million searches, and that only for few countries over one month of August (record isolation). I joined the pieces of search history beyond the defined cap of 3, more than 200,000 sets just for Germany and only in August (linkage). I also confirm that it is possible to perform inference on the dataset, for example concerning health, religion or political opinions. This moves us beyond data protection, and in the national security direction.
My report: "Alone among a thousand". (link below)
We just made all Macs touchscreen.
@sentientos_ai is a personal agent that lives on your Mac's notch. Every night, an on-device LLM understands your whole life so that you can just tap your notch to take over anything on your screen, with its own mouse cursor.
This magical experience also powers instantly replying to anyone in your own voice with your whole life in context.
Go experience the magic!
Kairox Systems (YC F26) is building autonomous drones that fly and swim.
Militaries around the world rely on unmanned systems for cross-domain air and sea missions, but they still need separate drones, separate operators, and separate logistics chains. Navy officers, special operations (SOF) operators, and researchers at MIT all told us the same thing: Aerial drones are fast but easy to detect. Underwater drones are stealthy but slow to get on target. At Kairox, we believe hybrid drones are the next evolution of unmanned platforms.
a16z's Jon Lai on why AI could shorten game development cycles and enable game studios to take riskier bets:
"One of the structural challenges in games is that it just takes a really long time to make these games, right? Grand Theft Auto 6 has been in production since 2013, 2014. It's a 12, 13-year development cycle."
"By any other software development standard, right? Taking 10-plus years to make HR software or CRM, it's kind of insane, right?"
"Let's see if we can get a 5 to 10-year dev cycle to, I don't know, 2 to 5 years. It's still going to take some time, but I think it'll just enable game developers to take more risk, have more at-bats, take more experiments, and just generally increase the richness and diversity of games that are created."
@Tocelot
AI is making games faster and cheaper to build, giving developers more room to take creative risks, revive overlooked genres, and let players reshape the worlds they play in.
In this conversation, a16z's Jon Lai and Anish Acharya join Elena Burger to discuss why human taste still determines what makes a game fun, how modding created some of gaming's biggest franchises, the opportunity for new distribution platforms, and games that adapt to whatever players choose to do.
0:00 Intro
0:57 How AI can now build a game
6:21 What shorter dev cycles unlock
10:08 Can AI make a game fun?
12:27 When creating is the game
14:01 Building a Habbo Hotel record shop
16:16 His son rebuilt Ghost of Tsushima in Minecraft
17:54 Mods: where billion-dollar franchises come from
20:36 Decompiling and reviving dead games
22:45 The renaissance of niche genres
25:51 Managing AI agents like an RTS game
28:02 The business of AI games
33:53 What excites and scares studios
35:16 The never-ending game
37:56 The games they'd finally buQild
YouTube: youtu.be/HDRy6XmE2RQ@VirtualElena@illscience@Tocelot
Today, we’re excited to introduce @luminai Topology.
In healthcare, every second can be the difference between life and death.
Cleveland Clinic used to take 3+ days to process a referral. Now, with Luminai Topology, referrals are processed the same day.
A 🧵on how:
Most of us open Google Sheets for budgets, trackers, and tables. With Gemini 4 Hydrogen, we wanted to see if it could paint a moving landscape instead 🍃
In Google @Antigravity, Hydrogen handled the coding to paint and animate thousands of cells — creating coastal trains, drifting clouds, and falling autumn leaves that recalculate frame by frame inside the spreadsheet.
(1/6) The biggest model upgrade to Search yet has landed. Gemini 4 Hydrogen is here, and brings cutting-edge intelligence to all Search users. It’s now the new default model in AI Mode & being used for your hardest questions in AI Overviews.
> Breaks down complicated, multi-part Qs & precisely follows your prompts
> Reasoning advancements improve query fan-out to show more links to the web
> Much more visual, scannable & highly tailored responses (with comparison tables, charts & more)
> Rich understanding of all supported languages
I’ve been testing the model behind the scenes and it’s really incredible. It’s one of my favorite models I’ve ever used, and can’t wait for you all to try it! Here are a bunch of examples I found helpful 👇
Announcing two new cyber products:
- Daybreak Defense Factory ()
- Daybreak Red Team (openai.com/daybreak/red-t…)
Both make it much easier to put frontier cyber capabilities to work, helping defend companies in new ways.
We're turning our lessons from defending OpenAI into practical tools for our customers. On Thursday, join us at Cyber at the Frontier, a live keynote, for our latest cyber announcements.
Working with our frontier models, we've identified and resolved important vulnerabilities in our own code and infrastructure. That experience is informing the tools and playbooks we're building to help other security teams strengthen their defenses.
@gdb shares what we've learned and why security teams should keep putting the latest models to work as capabilities improve.
Tune in on November 12 at 10 a.m. PT: openai.com/cyber-at-the-f…
🚀 Check out the latest @code updates
Bring Copilot CLI sessions into VS Code, manage extensions right from the Agents window, create custom chat layouts, and much more!
🔗 aka.ms/VSCode/146
Today we're giving the space industry its first universal refueling port.
OpenPort™ lets any spacecraft refuel any other. Two identical ports. Fuel flows either way.
Any company can license it for $1.
Satellites worth hundreds of millions are retired with empty tanks and working electronics. That isn't an engineering problem. It's a missing standard. Now it exists.
OpenPort is step one of the in-space refueling network we're building at @SpaceiumInc .
Build it to be refueled.
#SpaceTech#InSpaceRefueling
Ok this is a big deal. We’ve been testing Gemini 4 Hydrogen in early access with the Box AI Complex Work Eval, and it’s a serious jump for fast models.
Hydrogen is 5 percentage points better in accuracy than Gemini 3.8 Flash. That alone would be a great boost in capability, but possibly more importantly it did the tasks in less than half the time with about 12% fewer tokens on a typical task.
For this eval, we give the Box Agent Hydrogen as a model for a series of complex tasks that represent real world work with enterprise documents across finance, consulting, life sciences, and other industries. Hydrogen will be a great model for a lot of efficient but complex tasks across these fields.
Gemini team is definitely cooking. Hydrogen will be rolling out in the Box AI Studio shortly so any customer can build a custom agent with it.
Introducing Gemini 4 Hydrogen, an extremely capable and fast model built to bring Gemini 4 to everyone!
Available to all tiers of Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro + Ultra subscribers in Google Antigravity.
Introducing Gemini 4 Hydrogen–bringing cutting-edge intelligence with speed and efficiency to everyday coding and agentic enterprise work.
Learn more → goo.gle/4zTuFDK
Intelligent eyewear pre-orders are officially open! ✨
I’m immensely proud and humbled working with teams at @Google, @SamsungMobile, @_GentleMonster_ and @WarbyParker to bring these beautiful eyewear collections to life.
These lightweight glasses bring you hands-free help from @GeminiApp throughout your day. And these glasses are eyewear first, so @WarbyParker and @_GentleMonster_ have brought their iconic styles with a collection of designs, colorways, and sizes so there’s a style that’s perfect for you!
Pre-order your pair today: glasses.google and experience them in select Warby Parker, Gentle Monster and Google stores starting today. Shipping Dec 10th!
Read more about the full design collection that and how we are bringing helpful experiences to your everyday life: blog.google/innovation-and…
Along with our privacy protections for you and those around you: blog.google/innovation-and…#Gemini#IntelligentEyewear#AndroidXR
we are happy to share that we raised $8m from Peak XV, Fin Capital, Robinhood and some amazing angels.
when we started @agentcardhq we asked ourselves a question: are people willing to buy things with agents? not so many months later Instinct happened, Muse happened and what we thought was going to take years was taking place now.
agents buying things on behalf of humans is still an unsolved problem. using users’ cards is one part of it but making the purchase itself is another big one (we announced opencommerce.sh last week in an attempt to help solve this problem).
some months later we are still very much obsessed with this problem and with helping build this future for the next generations. it's a true privilege to work every day on problems with no simple solutions and to try to create clean and beautiful interfaces so that more companies can bring agents to every person on earth.
thanks to our investors for trusting us and to all of you for following this crazy ride. more cool things to come :)
- @PipeAbellos , @ValentinoConst6 , @pilisuagi and me
Lynyrd Skynyrd, Free Bird, Live, Oakland, 1977 - Remaster w/ Color & Gamma Corrections (Attempt 1) youtu.be/A4qwPYHkMTw
Billy Powell at the piano.
I was about to turn 16, between sophomore and junior years at Cubberley High School in (good) old Palo Alto.
Gemini 4 Hydrogen excels at knowledge work.
Based on over 6,000 real-world agentic sessions, Hydrogen (High) is #5 in Agent Arena overall. On knowledge-work tasks, it moves up to #4.
As a result, it’s ahead of three rival models that outrank it overall: Claude Opus 5.5 (Max), GPT-6 Astra (Max) and Claude Sonnet 5.5 (Max).
This means that @GoogleDeepMind's Hydrogen is not only the fastest model in the Agent Arena top 5, but also a powerful choice for real-world work.
Introducing Benchmark Router - one place to run public benchmarks on any model / your harness. At a 1/10th of the cost to run yourself*.
We have new LLM benchmarks launching... every 6 hours. Our goal with this platform is to give AI builders a unified place to run benchmarks without dealing with infra challenges - and understand how their agent stack-ranks.
Try us at benchmarkrouter.newmeasure.ai - pick a benchmark(s), a model(s), and a config. We run the official code, you get the metrics + full logs + failure analysis.
We're live with benchmarks across knowledge, coding, agentic, legal, finance, health, science and vision - and actively adding more, so would love for you to request them on the platform!
Lando heads out to find inspiration for his junior team’s playlist, with a little help from @GeminiApp on new intelligent eyewear from @_GentleMonster_.
So excited to share my YC launch!
@plek_ai turns an employee census into a bottom-up value creation plan in under a minute.
At a 3,600-person company, Plek Al found $30M in annual EBITDA levers, the same number a global PE advisory reached in 6 weeks
구글이 11월 9일부터 11일까지 TPU 8tx·8i Max, Gemini Robotics 2.5, AlphaProof Code, Gemini 4 Carbon, Gemini Decision 1, AlphaGenome 2 등을 연달아 출시했고, 오늘은 Hydrogen(Flash 모델)이나 Fable 5.2를 능가할 것으로 예상되는 Krypton의 조기 공개가 있을 수 있다는 내용이다.
well damn google is having one of its craziest release weeks ever:
nov 9 — tpu 8tx and 8i max
nov 10 — gemini robotics 2.5
nov 10 — alphaproof code
nov 10 — gemini 4 carbon
nov 11 — gemini decision 1
nov 11 — alphagenome 2
and today we might get either Hydrogen (Flash model) or a surprise early release of Krypton which is expected to surpass Fable 5.2
The world has made real progress on maternal mortality, with deaths falling by about 40 percent over the last two decades. But the pace of progress has slowed, and women are still dying from causes we know how to prevent. Getting the right tools and care to more mothers could change that.
Now rolling out for:
Developers: Build with the Gemini API in @GoogleAIStudio and Android Studio, and in Google @Antigravity for paid Google AI Pro and Ultra subscribers.
Enterprises: Deploy via Gemini for business and Google @Antigravity
Everyone: Access it in Google Search and the @Geminiapp model dropdown.
See the details → goo.gle/4zDCs8N
Introducing Gemini 4 Hydrogen
While Gemini 4 Carbon and Argon are built for deep, long horizon reasoning, Gemini 4 Hydrogen brings cutting-edge intelligence to everyone — with speed and efficiency for everyday coding and agentic workflows.
Surpassing Argon on the Artificial Analysis Intelligence Index at its default thinking level, at 2.5x the speed of Gemini 3.8 Flash, Hydrogen is engineered for high-velocity everyday work with:
— Multi-model orchestration & parallel agent workflows in Google @Antigravity
— Four tunable thinking levels for reasoning on demand (low to xhigh)
— Reliable harness execution for iterative coding & debugging
— Multimodal & embodied reasoning at scale
Start building with Gemini 4 Hydrogen today: with the Gemini API in @GoogleAIStudio and @AndroidStudio, client-side via @Firebase AI Logic, in Google @Antigravity for paid Google AI Pro and Ultra subscribers, and in Gemini for business.
Get the details:
Pumped to finally launch @mantlemachines to the world, an AI-native maintenance provider for America's industrial machines and robots.
Building machines and robots isn’t the bottleneck anymore, keeping them running is.
Today @GrahamShips and I change that.
Live across the US with multiple customers and 250+ technicians.
mantlemachines.com
Hydrogen brings major leaps in intelligence, speed, and reliability:
⚡ Cutting-edge intelligence: Surpasses Argon on the @ArtificialAnlys Intelligence Index at default high thinking (and climbs even higher at xhigh), while finishing tasks faster than competing models in its class.
💻 Cleaner coding & debugging: More than triples 3.8 Flash on Terminal-Bench 4.1, cuts fabricated claims by more than 5X, and adapts cleanly when a tool call fails or a search returns empty.
💼 Top marks on real-world work: Ranks #1 on Vals Tax Agent Bench, ties for #1 on Vals Legal Research, and leads on AutomationBench-AA, Multimodal Knowledge Agent (MKA), and Agents' Last Exam.
Never read an Airbnb listing again. Tell Jev what you want in plain English, and it reads 1,000 Cape Town listings and hands you the 10 best, scored out of 10, in about a second. Right inside Google Sheets 🏠
(trying something new with our demos!)
Today, @IndexLawHQ (YC F26) is launching Mike-1, an AI legal agent that understands your business, gets legal work done, and brings in exceptional human lawyers when you need them.
Give Mike a contract to negotiate. It reviews the agreement, explains the risks, prepares redlines, and helps get it ready for signature. You review and approve the work along the way.
When a matter needs a human lawyer, Mike brings in a vetted specialist with the documents and context already prepared. Your contracts, past negotiations, and company knowledge stay in one place to inform future work.
We’re starting with contracts and corporate work for startups and growing businesses.
We’ve already helped turn around contracts in under an hour and quoted legal work at up to 80% below traditional law firm alternatives.
That means helping businesses close deals faster and get legal support for decisions they might otherwise make alone.
Our ambition is to make the capabilities of a great law firm available as software.
Have a contract or corporate matter that needs moving forward? We'd love to help. Reach out to @qatindex at q@indexlaw.com or book a time here: cal.com/mike1/mike-1-s…
Spaceium (S24) is building a network of refueling stations in space, so spacecraft can refuel in orbit instead of being discarded when they run out of fuel.
Today, they're launching OpenPort, an open standard for spacecraft refueling that any space company can license for $1 and manufacture themselves.
In this fireside, @SpaceiumInc founders Ashi Dissanayake and Reza Fetanat join YC’s Brad Flora to explain how they’re turning spacecraft refueling from a custom engineering problem into shared infrastructure for the space economy. They share how they’ve built and tested orbital hardware at software-company speed, as they prepare to transfer xenon outside the ISS in 2027 and launch their first full refueling station in 2028.
00:00 — Building Gas Stations in Space
03:03 — Why Spacecraft Need a Refueling Standard
07:23 — Why OpenPort Costs $1
10:17 — Spaceium’s First Missions
14:41 — Building Hardware Like a Software Company
17:19 — How Spaceium Got Started
20:50 — The Hardest Part of Building a Space Company
22:25 — Making Refueling in Space Boring
Make a file with unique words found by no extant engine, use Google's API to index it, don't link to it from anywhere. Wait a day. Then query all the engines or APIs run by the other handles you tagged for the unique words, and see what happens.
ICYMI, .
I’m old enough to be the dad of half my YC batch. This is my second time around.
Here’s what I know now that I didn’t in W12.
Nobody knows what they’re doing. The first time, I was sure everyone else had it figured out. Their ideas sounded smarter. Their traction looked better. Then I watched batchmates pivot, rebuild and start over, just like we did. Now I know we’re all in the same boat, just trying to figure it out. Some people are better at hiding it.
An exit doesn’t give you the answers. Selling my first company to Dropbox made a few things easier. Investors take the meeting. People return your emails. But the product still has to work, and users still don’t care who you are.
Everyone says startups are hard. The first time, I nodded along and didn’t really believe it. Now I have the scars.
Consistency beats intensity. I’ve run every week for 200 weeks straight. Most of them I don’t feel like starting. Startups work the same way. You don’t win on the big days. You win by showing up every day.
Same nerves as 2012. Better shoes.
AI can help defend America's critical infrastructure as fast as attackers now operate.
Much of the equipment running power grids, water systems, and other essential services is decades old. Some of it can't be patched without risking an outage.
a16z's Anne Neuberger proposes a public-private defense plan focused on the few hundred operators and shared vendors whose failure could cascade nationally, across power, banking and payments, water, communications, and hospitals.
Super Intelligence-Era Critical Infrastructure Cyber Defense Plan: a16z.news/p/super-intell…
We cannot defend America’s infrastructure at human speed against attacks moving at machine speed.
In today’s WSJ, I make the case for using AI to protect the systems that keep our water flowing, hospitals operating and lights on.
Here’s the action plan: a16z.news/p/super-intell…
- Start with the few hundred operators and shared suppliers whose failure could cascade nationally.
-The White House must set outcome-based goals, like ensuring control systems aren't exposed to the Internet.
- Use AI to find and fix flaws. Contain what cannot be patched. Test whether essential services can keep running under attack.
- CEOs of critical infrastructure firms should be accountable for these outcome-based results.
- The White House, infrastructure operators, AI labs and technology providers each have a role. AI labs, for example, should work with critical infrastructure operators to build digital replicas so high risk patches can be tested offline.
AI can give defenders an advantage. We need to organize to use it.
wsj.com/opinion/cybera…
We’re betting on simulation to unlock the next era of robot learning.
Introducing @outworldlabs. We’re building a simulation platform for training and evaluating robot foundation models.
Collecting real-world data and testing robots is expensive and difficult to scale. Simulation offers a path forward, but it needs to reproduce real-world behavior and be easier to use.
We’re building a new physics engine for dexterous manipulation, with tools to create environments, reconstruct and augment real-world data, and scale training and evaluation in one platform.
Working on robot learning? We’d love to chat.
For developing 3D games and interactive apps, tracking down broken meshes, missing textures, or performance bottlenecks can stall your whole build. With Hydrogen’s spatial reasoning and surgical edits, you can use it as an agent that inspects your work and auto-fixes issues without breaking your master files.
Watch how the 3D Medic Managed Agent app built in @GoogleAIStudio uses Gemini 4 Hydrogen to inspect your 3D files, catch errors, and patch them before export. 🪄
There is no "pure data" company.
Data is what's given, often implicitly by user action, and it is not a commodity: it is non-rivalrous, perishable, non-fungible, non-depletable, ~zero marginal cost, etc.
Google isn't a data company without users who run its apps.
Use @Brave.
Scrape != Crawl.
Real search engines run global algorithms over very big data.
Search data from users isn't crawled. You can scrape SERPs or API results, but these decay. You need lead user signals to find fresh gold.
Economics and law come to the fore:
search.brave.com/ask?q=google+g…
Google has been named a Leader in the inaugural 2026 @Gartner_inc Magic Quadrant™ for Data Lakehouse Platforms!
We believe this validates our mission to deliver a borderless data foundation where teams can seamlessly run, manage, and scale their data estate.
Learn more: goo.gle/4AnMf2I
Orchestrator is rolling out today to Google AI Ultra subscribers (with additional models and plans supported in the future), and Enterprises can access it in Gemini for business.
Read the blog ↓
There are more tools than ever to help mothers survive and thrive. But these lifesaving breakthroughs will only save lives if they reach the people who need them in time:
Saving lives also means catching risks earlier in pregnancy. Pre-eclampsia, a dangerous rise in blood pressure, kills about 46,000 mothers a year. A new portable retinal scanner is helping rural clinics spot warning signs sooner:
Gemini 4 Hydrogen is rolling out starting today:
✨ Everyone: @GeminiApp, Google Search, and @GoogleAIStudio
✨ Google AI Pro & Ultra subscribers: @Antigravity, Gemini in Chrome, and Gemini in Google Sheets
✨ Developers & enterprises: All tiers of the Gemini API in @GoogleAIStudio, @AndroidStudio, and Gemini for business
Learn more ↓
Retell just crossed $100M in revenue run rate.
$10M to $100M in 17 months. Profitable. $4.6M raised.
From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us.
Our bet: the future of the contact center belongs to the teams who run it.
Proud of our team. Grateful to every customer who believed in us.
I was invited to Y Combinator’s first-ever Startup School in Asia! Absolutely love the energy in Taipei🍵🧋
So grateful for the opportunity to connect with so many key leaders and ecosystem builders in Taipei! @ycombinator
:-(
The U.S. may do something as bad too. The dilemma is undereducated deciders' decrees to "share data", based on "data is the new oil" stinking thinking.
This means either anonymizing so much the data is of little value (better), or id everyone (bad).
There is no "pure data" company.
Data is what's given, often implicitly by user action, and it is not a commodity: it is non-rivalrous, perishable, non-fungible, non-depletable, ~zero marginal cost, etc.
Google isn't a data company without users who run its apps.
Use @Brave.
There is no "pure data" company.
Data is what's given, often implicitly by user action, and it is not a commodity: it is non-rivalrous, perishable, non-fungible, non-depletable, ~zero marginal cost, etc.
Google isn't a data company without users who run its apps.
Use @Brave.
3/Search and scraping apps grew 160%, adding more calls than any other category. Agents check the web before they act:
@ExaAILabs +120%
@tavilyai +160%
@firecrawl +95%
@perplexity_ai is new on the list and @brave grew 70%.
If you’re building websites or storefronts, Hydrogen’s fast generation and first-pass coding accuracy let you go from a single prompt to a working frontend in minutes. Using our Gemini frontend design skill in Google @Antigravity, Hydrogen built this e-commerce store complete with Nano Banana product photography and smooth scroll effects. 🛒
Get the skill here: goo.gle/4BUKVWG
@grok write a script that takes a word and a letter. Return a count of the number of that letter in that word. Run that script over the following word/letter pairs:
seventeen/e
railroad/r
bookkeeper/k and o and e.
Not responding to the canoe of handles you filled :-P.
We don't scrape. Scrape denotes querying bigs from Google down.
All search engines crawl, for a lighter touch on many sites, not heavy on SERPs of big few.
Brave Search respects robots.txt and we do not blind-crawl either.
100% is asymptotic. I'm typically happy with high 90s. I allow CC < 10 for agents -- their short term memory is better than ours. For humans I prefer < 4.
@grok in the normal spelling of the word "seventeen", how many time does the letter 'e' appear?
