The Rundown AI

A big day for Google! Gemini 4 Argon is here, taking first place on 13 of the 19 benchmarks Google published against GPT-6 Astra and Claude Opus 5.5. A few standout numbers: - DeepSWE (long real-world coding tasks) – 77.9%, vs. 74.2% for Opus 5.5 and 74.1% for Astra - Vals Index (real work in finance, law, tax, coding) – 68.9%, vs. 67.0% for Opus 5.5 and 63.1% for Astra - Output limit is 1M tokens, up from 64K Argon also leads on tests for vibe coding, long context, and reading charts and long videos. The release is limited to trusted cyber defenders in Google's Fairwind Program to start.

Apple Vision Pro's most practical use yet might be in the operating room. A Duke Health surgeon wore it through a hip surgery, with a live camera feed from inside the patient's joint floating in view:

Top stories in AI today: - OpenAI puts a 24/7 agent inside ChatGPT - Anthropic's doom-heavy IPO filing leaks - Get started with ChatGPT dot, OpenAI's new agent - The U.S. federal government gets an AI front door

Google just brought AI watermarking into the physical world. DeepMind embedded detectable signatures in AI-designed proteins, and they still worked after being made in a lab. The goal is to let DNA makers check that a design came from an AI model with safety guardrails:

Top stories in AI today: - Gemini 4 Argon looks to return Google to the frontier - Nate's Notebook: When judgment gets cheap - How (and why) to set up Meta’s Muse Desktop - Behind the scenes at the White House’s big AI day

Impressive demo from Tavus of its new 'Human Interaction Model' named Griffin, which the company says passes the video Turing test. 48% of people who talked to it live thought it was a real human, up from under 3% for earlier models:

Top stories in robotics today: - DoorDash builds its own delivery drone - Dyna’s robot handles the entire laundry stack - ETH teaches disembodied robot hand to crawl - MIT puts living muscle into a paper-thin robot

Top stories in AI today: - Atomic Machines’ AI turns code into tiny machines - The Rundown Roundtable: Our AI use cases - Turn a product photo into three matching ads - Microsoft CEO: Treat AI like it’s already been hacked

NEW: U.S. President Donald Trump, on Truth Social: "The White House considers anyone that uses the term, 'Artificial Intelligence,' as opposed to the highly accepted new and more accurate term, 'Super Intelligence,' THE ENEMY!"

Artificial Analysis just rebuilt its "Intelligence Index" AI leaderboard, with 60% of it now running on private tests models can’t study for. The new standings: 1. Claude Opus 5.5 – 58, still topping the charts 2. GPT-6 Astra – 57 3. GPT-6.1 Sol – 55 4. Claude Sonnet 5.5 – 54 5. GPT-6.1 Terra – 51, ahead of Gemini 4 Argon (50) and Claude Fable 5.1 (49) at ~1/13 of Fable’s cost Kimi K3.1 Pro, Moonshot’s model with open weights due Nov. 4, fell from 55 to 46 under the new testing, dropping from just 3 points behind the top spot to 12. The best open models available today, DeepSeek V4.1 Flash and GLM-5.3, score 32, below Claude Haiku 5.5 (36).

NEW: Meta released Muse Wonder (codenamed Watermelon), its new family of AI models, with scores that push the company to the frontier. The top version, Wonder Full, scores 56 on Artificial Analysis’ Intelligence Index, behind only Claude Opus 5.5 at 58. It also leads rivals on: - Legal work (Harvey LAB): 13.1 vs. 8.6 for GPT-6 Astra - Long coding projects (SWE-Marathon): 57.5 vs. 46.3 for Opus 5.5 - Coding tasks (DeepSWE): 79.0 vs. 74.2 for Opus 5.5 - Remembering context (MemoryBench): 85.5 vs. 81.5 for Opus 5.5 Muse Wonder is live today in the Meta Model API and Muse Code, with Wonder Full open to paying Muse Code users.

Top stories in AI today: - Tavus’ Griffin passes for human on video calls - Rowan’s Corner: My recurring AI performance review - Is ChatGPT ‘Dots’ worth upgrading to Pro for? - Google’s AI watermarks cross into the lab