🎬 OpenAI pulls the plug on Sora, bets everything on 'Spud' 🎬
The company sacrifices its video flagship to power its most ambitious model yet

In a bold strategic pivot, OpenAI has completely shut down Sora — its app, API, and all video generation features inside ChatGPT are gone. The freed compute is being redirected to the company's next model, codenamed "Spud," which CEO Sam Altman reportedly told employees could "really accelerate the economy." The entire Sora team is being reassigned to robotics and "world simulation" projects, signaling that compute allocation — not product breadth — now drives OpenAI's decision-making. The move leaves Google as one of the only major players still pushing AI video at scale, and confirms OpenAI views foundational model capability as more valuable than competing in the increasingly crowded generative video market.

🤖 Jensen Huang says: "we achieved AGI" 🤖
Nvidia's CEO redefines the AI goalpost based on what today's systems can already do

In a wide-ranging conversation on the Lex Fridman Podcast, Nvidia CEO Jensen Huang stated what few industry leaders have dared to: "I think we've achieved AGI." Huang isn't invoking science fiction — he's pointing at how current AI systems already automate complex tasks, write production code, manage workflows, and operate with growing independence. The remark effectively shifts the AGI goalpost from theoretical to practical, which changes how the entire industry frames its competitive roadmap. The definition of AGI remains deeply contested, and today's models still stumble on tasks requiring robust reasoning and common sense. But coming from the CEO whose company powers most of the world's AI infrastructure, the statement carries serious strategic weight.

📊 ARC-AGI-3 resets the scoreboard and humbles every frontier model 📊
Humans solve 100% while GPT-5.4, Gemini, Claude, and Grok score below 0.4%

The ARC Prize Foundation dropped ARC-AGI-3, a new interactive reasoning benchmark designed to measure agentic intelligence — and the results are humbling. Gemini 3.1 Pro scored 0.37%, GPT-5.4 hit 0.26%, Claude Opus 4.6 managed 0.25%, and Grok 4.2 scored a flat 0%. Meanwhile, humans solved 100% of environments on their first attempt with no prior training. Labs had invested millions training on the previous ARC-AGI-2 version, pushing scores from 3% to roughly 50% in under a year — but this new release wipes those gains entirely. A $1 million prize backs the challenge, spotlighting the persistent chasm between sophisticated pattern-matching and genuine reasoning.

INTERESTING TO KNOW

🔬 OpenAI maps the path to a fully autonomous AI researcher by 2028 🔬

OpenAI revealed it's building a multi-agent system it calls the "AI researcher" — designed to handle complex tasks across math, science, coding, and policy with minimal human oversight. An intermediate version, dubbed the "AI research intern" for smaller multi-day tasks, is expected by September 2026, with a fully autonomous system targeted for 2028. The architecture combines Codex, reasoning models, and monitoring systems with sandboxed environments for safety. In parallel, the company plans to nearly double its headcount to 8,000 employees by year-end — hiring across product, engineering, research, and sales.

💰 Anthropic crosses $19B in annual revenue as Apple secretly runs on Claude 💰

Anthropic is now generating $19 billion in annualized revenue — a staggering 20x increase in roughly 15 months — driven largely by enterprise adoption and Claude Code, which accounts for about 4% of all public GitHub commits. Meanwhile, Bloomberg reports that Apple internally "runs its business on Anthropic" for product development, hosting Claude on its own servers for code generation, prototype testing, and security automation — even as it publicly partners with Google Gemini for Siri. Eight of the ten largest Fortune 10 companies are now Anthropic customers.