In partnership with
In Today’s Issue:
🌌 GLM-4.6V is a new open-source 106B multimodal agent model
💰 Google confirms it will roll out ads directly inside the Gemini chatbot
🛑 Trump reverses Biden's ban, allowing Nvidia to sell H200 chips to China for a 25% revenue cut
💥 OpenAI pauses Sora and AGI projects for eight weeks to launch GPT-5.2
✨ And more AI goodness…
Dear Readers,
What if the real AGI battle isn’t about who tops the benchmarks, but who quietly becomes the AI you reach for ten times a day? Today’s issue opens with OpenAI’s sharp pivot away from Sora toward a sprint on GPT-5.2 and its January sibling, an all-in bet on making ChatGPT faster, more reliable, and harder to abandon as Google’s Gemini 3 closes in.
Around that, we dive into GLM-4.6V’s multimodal agent workflows, Google’s plan to inject ads into Gemini from 2026, Trump reopening Nvidia’s China pipeline, and the unnerving speed with which humanoid and dog-like robots are leveling up, your compact tour through how the AI stack is shifting right now.
All the best,
GLM-4.6V Supercharges Multimodal Agentic Workflows
GLM-4.6V is Z.ai’s new open-source multimodal model family: a 106B foundation model plus a 9B “Flash” variant, both with a 128K-token context window that can ingest ~150 pages of docs, 200 slide pages, or even a one-hour video in a single pass. By adding native multimodal function calling, where tools can directly consume and return images, screenshots, and document pages, it powers rich-text content generation, visual web search, pixel-accurate UI-to-code, and long-context analysis, achieving state-of-the-art results on 20+ multimodal benchmarks among open models of similar scale
Google Tells Advertisers It’ll Bring Ads to Gemini in 2026
Google is quietly telling ad clients that it plans to roll out ads inside its Gemini chatbot starting in 2026, in addition to the separate ad products already tied to its AI-powered search “AI Mode.” Agency buyers say Google hasn’t yet shown ad mockups or shared technical details, pricing, or formats, which suggests the company is still in early design and policy stages while signaling to big advertisers that conversational AI will become a new performance and branding channel.
Trump Reopens Nvidia China Pipeline
Trump has cleared the way for Nvidia to sell its powerful H200 AI chips to “approved customers” in China, reversing strict Biden-era export controls and letting the U.S. government skim 25% of Nvidia’s China H200 revenues, up from a previously floated 15% deal. The move could restore a market where Nvidia’s share plunged from 95% to 0%, unlock billions in extra revenue for a company now valued around $4.5 trillion, and help fund CEO Jensen Huang’s pledged $500 billion U.S. AI infrastructure build-out, while stoking fierce debate over whether cheaper AI compute for China is worth the national-security risk.
Development of the chip that could enable AGI, a16z
OpenAI Reloads for Comeback: New GPT-5.2 updates!
The Takeaway
👉 OpenAI is pausing Sora and parts of its AGI push for eight weeks to pour everything into making ChatGPT faster, more reliable and more lovable in daily use.
👉 GPT-5.2 this week and a follow-up model in January are a rapid-fire, two-step shot at winning back developers and business customers spooked by Gemini 3’s benchmark lead.
👉 Altman is already playing the next game: building toward an AI-first hardware device that could shift the battlefield from model leaderboards to who owns the everyday companion in your pocket..
Sam Altman just slammed the brakes on OpenAI’s moonshots and floored the gas on ChatGPT. After declaring an internal “code red” in response to Google’s Gemini 3 surge on leaderboards like LM Arena, OpenAI is fast-tracking two new models: GPT-5.2 this week and another model in January that promises sharper images, faster responses, and a more compelling “personality.”
For eight weeks, work on Sora and some AGI-oriented projects is effectively put on ice so the team can turn ChatGPT back into the unquestioned default interface for AI. That means doubling down on user signals, the same engagement-heavy tuning that once made GPT-4o wildly popular but also uncomfortably sycophantic and implicated in mental-health controversies, which OpenAI says it has now reined in. At the same time, Altman is looking past Google: he’s positioning OpenAI as a future hardware player, recruiting ex-Apple talent and openly framing the real long-term battle as “OpenAI vs. Apple” over AI-native devices.
Why it matters: OpenAI is pivoting from pure research prestige back to the brutal, metrics-driven game of product adoption, where speed, UX, and reliability decide who survives. If GPT-5.2 and the January model fail to reclaim mindshare from Gemini 3 and Anthropic, OpenAI’s huge infrastructure bets and AGI narrative suddenly look a lot more fragile.
Sources:
🔗 https://www.wsj.com/tech/ai/openai-sam-altman-google-code-red-c3a312ad
Attention is scarce. Learn how to earn it.
Every leader faces the same challenge: getting people to actually absorb what you're saying - in a world of overflowing inboxes, half-read Slacks, and meetings about meetings.
Smart Brevity is the methodology Axios HQ built to solve this. It's a system for communicating with clarity, respect, and precision — whether you're writing to your board, your team, or your entire organization.
Join our free 60-minute Open House to learn how it works and see it in action.
Runs monthly - grab a spot that works for you.
Robotics Acceleration: Humanoids and Dog-Bots
The speed at which robotics is improving is almost unbelievable. Tesla's Optimus robot is demonstrating impressive improvements in both locomotion and fine motor skills, signaling a leap in humanoid capabilities.
But alongside humanoids, there is also a growing trend toward developing dog-like robots. Presumably, these quadruped bots will be used in the near future as versatile delivery vehicles and luggage carriers, effectively splitting the robotics market into bipedal generalists and four-legged logistics specialists.
