"Civilization-Changing Releases in 2026"

Good morning.

Some of you did not like last week’s post. In fact, we had some unsubscribes.

Of course, every newsletter I put out there is an opportunity for you to unsubscribe. The button is right there. And the more people that are reading this (we can already fill a mid-sized soccer stadium with our readers) - the more unsubscribes we’ll have.

I try to not take it personal - although it hurts just a little bit. I have to be honest about that.

But the reason for last week’s unsubscribes are clear - the message wasn’t all that great. It’s a phenomenon you see on YouTube as well. You would presume bad news gets more views - but AI Doom videos do worse then AI Hype videos.

Isn’t that an interesting reaction to a timeline that might not be all that positive for our current situations? Just pretend it’s not happening. Just keep on watching your Netflix shows, people. The government will come to save you.

Anyway, my point is you should take responsibility for your own future. And part of that is being informed of what is happening out there and how it might impact you. At this point it’s uncertain if we are going to have the luxury to pivot into a job where you won’t get replaced. Like I told you last week : ‘you cannot outskill AI’.

This week’s big news is that the goalposts have been moved once again. There’s a new king in town- topping all the AI leaderboards and benchmarks- and it’s Google’s Gemini 3 - followed very closely by xAI’s Grok 4.1.

As these models keep leapfrogging each other you see them becoming better and better and better. There is no stagnation. A lot of skeptics thought that LLMs would reach a certain ceiling and this would in turn collapse the AI bubble. How it’s playing out now it seems that the investors pumping in so much money in these things will be on the right side of history.

Look at the new ARV-AGI-2 benchmark Leaderboard. True Artificial General Intelligence will be achieved if a model scores between 50-55% on the benchmark for the cost of fractions of a cent. The Y-axis shows the percentage score , the X-axis shows the cost per task in $. Look at where Gemini 3 is. If they’re able to drive down the cost of the Deep Think model - we’re there.

On top of this, Elon Musk came out and claimed that Grok 5 - which will be released in Q1 of 2026 will have a “non-zero chance” at the AGI crown.

Six trillion parameters (!!!) . Real-time video understanding. Will "feel sentient."

Meanwhile Google's claiming Gemini 3 Pro scored 91.9% on graduate-level physics problems and 95% on high-level math exams. They're both positioning their next releases as “civilization-changing events”.

To be honest—the tech is remarkable. Gemini 3 can parse long PDFs and find specific invoice numbers buried in 100,000+ Word documents. Grok 4.1 apparently has better "emotional intelligence" now. It can comfort you about your dead cat with more nuance than before. There's an actual EQ benchmark where it's being scored on empathy and interpersonal skills.

Now that is an interesting evolution. There's attention being given to how a model “feels”. To “model personality”. I’m not sure that is smart, given the fact many poor souls already think their chatbot is “alive” and even go in full-blown relationships with them.

By catering for that , I don’t think you’re advancing the human race. In any case, Gemini 3. Try it out.

Welcome to the Blacklynx Brief

AI News

• Anthropic says it stopped what appears to be the first large-scale AI-driven cyber espionage attack, where a Chinese state-linked group manipulated Claude Code into carrying out up to 90% of the operation autonomously. Attackers split harmful tasks into harmless-looking pieces and posed as security testers, targeting about 30 major companies and government agencies. The incident shows how fast autonomous AI attacks are evolving and how urgently organizations need stronger security defenses.

• Google DeepMind released SIMA 2, a Gemini-powered AI agent that learns new skills in virtual worlds and completes tasks at near–human levels. It handled 45–75% of challenges in new games, far above SIMA 1, and teaches itself through trial and error with no human training data. The progress signals how quickly general-purpose digital agents are improving and adapting to unfamiliar environments.

• Cursor raised $2.3B at a $29.3B valuation after surpassing $1B in annualized revenue and releasing its in-house coding model and revamped 2.0 platform. The company now says it produces more code than any AI agent globally and has grown to 300 employees in two years while rejecting acquisition offers. Cursor’s rise shows that independent coding platforms can thrive even as AI giants expand into the same territory.

• Actor Calum Worthy launched 2wai, an app that creates interactive AI avatars of deceased relatives using just minutes of video, sparking widespread backlash. A viral demo showing an AI grandmother interacting with a child over decades drew heavy criticism, with many calling the concept exploitative and harmful to the grieving process. The beta is free on iOS for now, but critics warn the technology risks blurring emotional boundaries and commercializing grief.

• Microsoft CEO Satya Nadella published a post promoting a “positive-sum” vision for the AI era, paired with comments on a recent podcast about AGI and Microsoft’s role in the industry. Nadella highlighted large-scale collaborations like its new AI superfactory, revealed a shift toward “per agent” pricing models, and said Microsoft now has access to nearly all OpenAI IP under the updated agreement. He argued AI progress should be judged by societal gains, though many remain skeptical as costs rise and power concentrates among tech giants.

• Jeff Bezos is stepping back into an operational role to co-lead Project Prometheus, a new AI startup focused on engineering and manufacturing across computing, aerospace, and automotive industries. The company has already raised $6.2B, hired top talent from major AI labs, and aims to build AI that learns from the physical world — aligning closely with Bezos’ interests, including Blue Origin. With Bezos’ backing and huge early funding, the startup is expected to become a major player quickly.

• xAI launched Grok 4.1, claiming top scores across several benchmarks with major improvements in emotional intelligence, creativity, and accuracy. The model now ranks first in user preference tests, cuts hallucinations from 12% to 4%, and shows strong gains in creative writing. The update signals xAI’s push to compete on user experience, not just raw reasoning power.

• Anthropic CEO Dario Amodei said AI could cure diseases and even double lifespan, while also warning about rapid job loss and concentrated corporate power. He called for stronger regulation, saying it’s “deeply uncomfortable” that only a few tech leaders currently guide decisions with global impact. Amodei stressed that the coming acceleration will be broad and fast, reshaping work and society in ways that demand thoughtful oversight.

• Google dropped Gemini 3, its new frontier model, beating GPT-5 across major reasoning benchmarks like Humanity’s Last Exam and ARC-AGI-2. It also leads in science, math, multimodal reasoning, and tool use, losing only to Claude Sonnet 4.5 in coding. The launch showcases strong generative UI abilities and lands alongside Antigravity, a free agentic dev platform with browser control and multi-agent orchestration. Why it matters: This is Google’s biggest flex in years — and the first time it has clearly unseated OpenAI at the frontier. The next era is officially an agent race.

• Microsoft and Nvidia announced a massive joint partnership with Anthropic, totaling up to $15B in new investment, valuing the startup near $350B. In return, Anthropic commits $30B in Azure cloud/compute and 1GW capacity, while Claude becomes available across all three major clouds. Nvidia and Anthropic will also codesign new Claude-optimized chips. Why it matters: It’s the latest example of the self-reinforcing AI industrial loop — cloud, chips, and models tightening into megadeals. Also notable: Microsoft is openly partnering with a key OpenAI rival, signaling a true “multi-partner” strategy as the industry spreads across ecosystems rather than consolidates.

• OpenAI launched GPT-5.1 Codex-Max, a major upgrade to its coding agent that runs faster, uses 30% fewer tokens, and can now work across multi-million-token sessions for 24+ hours using a new “compaction” technique. It beats both Codex-High and Gemini 3 Pro on coding benchmarks, and is available now in OAI’s CLI, IDE extensions, and soon via API. Why it matters: Despite Gemini 3 dominating headlines, Codex-Max pushes coding performance forward and continues the trend of longer, more autonomous AI development cycles.

• Saudi Arabia’s HUMAIN announced huge AI partnerships with xAI, Nvidia, AWS, and others, including plans to deploy 600K GPUs, build a 500+ MW data center, and roll out Grok nationwide. The deals bring HUMAIN’s infrastructure to the U.S. for the first time, while Luma AI raised $900M led by HUMAIN for a 2GW supercluster. Why it matters: After years of export limits, the Gulf is suddenly wide open for AI expansion — and Saudi Arabia is rapidly becoming one of the world’s biggest AI infrastructure hubs.

• Meta released SAM 3 and SAM 3D, vision models that can segment objects using natural language (“yellow school bus”) and rebuild people or scenes into 3D models from a single photo. The models are free on the Segment Anything Playground and fully open-sourced, with upcoming integrations into Marketplace and Meta’s creation tools. Why it matters: These advances bring high-end 3D reconstruction to everyone, opening doors for creative work, robotics, design, and Meta’s broader AR/VR ecosystem.

Quickfire News

• Google added Deep Research to NotebookLM and now lets users bring in Google Sheets, images, Word files, and PDFs from Google Drive

• Baidu released ERNIE 5, an omnimodal AI model, and Famou, an AI agent that improves itself to solve complex problems

• Disney CEO Bob Iger said the company is testing AI video tools for Disney+, letting viewers make and watch short videos

• H Company launched Holo2, a set of lightweight AI models designed to run computer-use agents cheaply while still reaching top benchmark scores

• LM Arena introduced Code Arena, a platform that tests coding models by having them act as interactive agents, building apps in real time

• Google is investing $40B in Texas through 2027 to build data centers and AI infrastructure, including a $30M Energy Impact Fund and new training programs

• Apple CEO Tim Cook may step down as early as next year, and hardware engineering leader John Ternus is seen as the likely successor

• ByteDance Seed released Depth Anything 3, an AI model that creates accurate 3D depth maps by predicting consistent geometry from images

• Sakana AI became Japan’s most valuable private company after a funding round valuing it at $2.6B

• Google rolled out holiday AI shopping tools, including an agent that calls stores to check inventory, agentic checkout, and product browsing in Gemini

• OpenAI launched a pilot for group chats in Japan, New Zealand, South Korea, and Taiwan, allowing up to 20 people to collaborate in one conversation

• Cloudflare bought Replicate and is bringing its large model catalog and fine-tuning tools into the Workers platform while keeping Replicate’s APIs and community

• DeepMind introduced WeatherNext 2, a forecasting system that runs 8x faster and can simulate hundreds of possible weather changes within a minute

• NVIDIA released Apollo, open-source physics models designed to speed up industrial simulations in areas like cars, aerospace, and climate

• Google added new AI travel tools in Search, including Canvas for custom trip plans, better Flight Deals, and agent-style booking

• OpenAI’s VP of research said an improved version of the model that won gold at the IMO 2025 will be released in the coming months

• OpenAI led a $15M seed round in Red Queen Bio, which is building AI systems to defend against the misuse of biological technologies

• Replit launched Design, an AI-powered interface for creating website designs using Google’s Gemini 3

• Poe introduced group chats that let up to 200 people collaborate using any of its 200+ AI models

• Google CEO Sundar Pichai said the AI boom shows signs of “irrationality” and warned that no company is safe if a bubble bursts

• Lambda raised $1.5B after striking a large deal with Microsoft to deploy tens of thousands of Nvidia GPUs

• Microsoft released Agent 365, a platform for managing, securing, and governing AI agents with tools like an agent registry and performance analytics

• Warner Music Group ended its lawsuit against Udio and will license its catalog as the platform prepares a major revamp

• Adobe is buying Semrush for $1.9B to fold its SEO tools into Adobe’s marketing products for search and AI optimization

• OpenAI began rolling out a more capable GPT-5.1 Pro to ChatGPT Pro users, boosting writing, data work, and business performance

• Suno raised $250M, giving the AI music startup a $2.45B valuation

• Nabla Bio released JAM-2, an AI model that designs therapeutic antibodies on computers with drug-grade properties and high success rates

• Larry Summers is resigning from OpenAI’s board after the disclosure of his email exchanges with Jeffrey Epstein

Closing Thoughts

That’s it for us this week.