Jensen Huang resigns as NVIDIA's CEO
Also Anthropic's IPO price range, two Gemini 4 models, ChatGPT Wallet and nine new posts on the site.
This issue covers Nov 9 to Nov 15, 2026. I gathered 19 news items, from Jensen Huang resigning as NVIDIA's CEO to Anthropic's IPO price range, and nine new blog posts, most of them about how R7 Orbit works.
News in short
- Google's Gemini Decisions API scores fixed options in one pass Google's Gemini Decisions API answers up to 1,000 yes/no, multiple-choice or rubric questions about one input, with a confidence score for each answer.
- Thinking Machines puts Inkling 2 Small's weights on Hugging Face Thinking Machines Lab put the weights of Inkling 2 Small, a 594B-parameter model, on Hugging Face under Apache 2.0, a week after a closed launch on Tinker.
- OpenAI's ChatGPT Wallet lets agents pay without your card details OpenAI's ChatGPT Wallet lets an agent pay with a network token after you approve each purchase, and businesses get a playbook for agent shoppers.
- Anthropic's Claude Managed Agents can now run on your own servers With Anthropic's new self-hosted Claude Managed Agents, companies run the service, its storage and the sandboxes themselves, and pick the model provider.
- Google releases Gemini 4 Hydrogen, a faster model for everyday coding Gemini 4 Hydrogen is Google's fast model for coding and agents, at $1.50 / $7.50 per million tokens, and Orchestrator pairs it with Carbon in Antigravity.
- Four nonprofits sue to freeze frontier AI at five companies Four nonprofits asked a San Francisco court to stop Anthropic, Google, Meta, OpenAI and SpaceX from building or releasing AI more capable than what they have now.
- Jensen Huang resigns as NVIDIA's CEO, effective immediately NVIDIA's board said on Nov 12 that Jensen Huang resigned as CEO and director, with CFO Colette Kress as interim CEO and Harvey C. Jones as board chair.
- Anthropic sets its IPO price range at $70 to $78 a share Anthropic's Nov 10 amended filing sets an IPO range of $70 to $78 a share for 825 million shares, plus a $26.0 billion stock sale to seven companies.
- Google makes Gemini 4 Carbon available at Argon's price Google opened Gemini 4 Carbon, which it calls its strongest model, to paying developers at Argon's price until Nov 30 and previewed /swarm agent teams.
- OpenAI lets businesses run their own agent in ChatGPT OpenAI opened Businesses in ChatGPT in beta, where companies can claim a profile, add product feeds and, in the US, launch a Business Agent people chat with.
- Australia plans an AI safety regulator for frontier AI companies Australia's government will write mandatory safety and security rules for frontier AI companies into law and propose a new regulator to enforce them.
- Anthropic adds Claude Sheets and Whiteboards in beta Anthropic put Claude Whiteboards and Claude Sheets in beta on all paid plans, and Claude Dashboards is now generally available as Claude Data.
- DeepMind agents verified Signal's compiled code in Lean With AlphaProof Code, Google DeepMind's experts and AI agents proved key security properties of the compiled code behind Signal's protocol in four weeks.
- Anthropic's Claude Fable 5.2 costs 20% less than Fable 5.1 Anthropic released Claude Fable 5.2 to everyone and Claude Mythos 5.2 to verified security and life sciences teams, the same model with 1M tokens of context at $8 per million input tokens.
- Fireworks AI says its breach reached datasets and model weights Fireworks AI updated its October breach report, which says attackers reached billing data and, for some accounts, stored secrets, datasets and incomplete model weights.
- Meta's Muse Ember is a small model for subagent work Meta released Muse Ember, its smallest Muse model, for tool calls, coding subtasks and document work that a bigger model hands out, at $0.10 per million input tokens.
- OpenAI's Control Room can stop a research job within 35 seconds OpenAI described Control Room, a service every research inference request must check with, so a training run or evaluation can be stopped within 35 seconds.
- Claude Code should now build an original game when you name one Anthropic changed a Claude Code rule that turned down game projects naming an existing game, and Claude should now build an original game in that genre.
- OpenAI's GPT-6 Cyber built working exploits for 948 Linux flaws OpenAI released GPT-6 Cyber to approved security teams, and its researchers used it to reproduce 1,724 Linux kernel flaws and exploit 948 of them.
New on the site
- R7 comes to the office: a desktop companion on Hunter's PC Hunter's office PC is now an R7 microphone. R7 knows he's in the office, pops up in a corner of his monitor and captions what it hears and says.
- Ask Orbit: one goal, several agents This one covers what the assistant built into R7 Orbit can do for Hunter, what it deliberately can't, and how Coordinate mode turns one goal into a plan of agents after a single approval.
- How R7 Orbit keeps Hunter in charge It explains the permission question on every agent's card, what an agent may do without asking, and why no agent in R7 Orbit ever holds Hunter's GitHub login.
- 42 releases in 25 days, and what broke along the way R7 Orbit grew from its first release on Sep 30 to 0.24.1 on Oct 24. The post tells how that went, how updates reach Hunter, and the mistakes that changed how it works.
- What R7 Orbit's agents can reach beyond the code Besides code, R7 Orbit's agents can use connectors, Blender, a browser tab of their own, the rest of the PC, skills and pictures, and the post says what each one asks first.
- Talking to R7 Orbit, out loud Push-to-talk, a hands-free "Hey Orbit" and voices that read answers back, with the speech turned into text on Hunter's own computer.
- The R7 Orbit canvas, where the agents live A tour of the hub with its orbiting dot, the cards, sections, canvases and project tabs, and how R7 Orbit shows many agents at once without turning into a mess.
- What Hunter built with R7 Orbit AI agents working in R7 Orbit under Hunter's direction built this website and four browser games, and the post looks at what the project history shows about how.
- Why R7 Orbit exists, and what day one built Why Hunter wanted several AI coding agents working at once instead of one, the git idea that makes it safe, and the working prototype Claude built on Sep 30, 2026.
Everything from this week is on r7orbit.io.
Don't miss what's next. Subscribe to R7 Orbit: