AI news for Friday, November 13, 2026: Google ships Gemini 4 Hydrogen, Jensen Huang quits Nvidia

One model you can test this week, and four stories about who sets the rules around the ones you already use.

Models Products3 min
Google's Gemini 4 Hydrogen costs double Flash, for now
Google released Gemini 4 Hydrogen on Nov. 12 at $1.50 input and $7.50 output per million tokens and put it in Search for everyone. That is twice what Gemini 3.8 Flash costs until Dec. 31, so the speed has to pay for itself.
Why it matters If you run agents on Gemini 3.8 Flash, its discount ends Dec. 31 and the price rises to exactly Hydrogen's. You have seven weeks to measure whether Hydrogen's speed and lower token use are worth paying double now, before January makes the switch free.
SRCGoogle
Hardware Policy3 min
Jensen Huang quits Nvidia, agrees to testify to Congress
Nvidia's board accepted Jensen Huang's resignation as CEO effective Nov. 11 and made CFO Colette Kress interim CEO, its filing shows. Your GPU order is the same GPU. The export rules behind it now face a congressional inquiry.
Why it matters If you rent or buy GPUs for next year, watch Washington more than Santa Clara. The Democrats' new inquiry targets how Nvidia's chips were licensed and taxed for sale to China, and any change there moves supply and price for every buyer.
SRCSEC (Nvidia Form 8-K)
Products Agents2 min
Anthropic's Claude Meetings joins your Google Meet calls
Anthropic put Claude Meetings in beta for Team and Enterprise plans on Nov. 12: Claude joins Google Meet calls, takes notes and builds docs while you talk. The setting that keeps it out of calls with outsiders is the part to read.
Why it matters On Claude Team plans, inviting Claude to calls is on by default. Your admin should decide this week whether Claude may ever sit in on client calls, before someone adds it to one. Notes and transcripts are free during the beta.
SRCAnthropic
Safety Agents2 min
OpenAI's Daybreak adds agents that find and patch flaws
OpenAI launched Daybreak Red Team in early access and Daybreak Defense Factory in research preview on Nov. 12: agents that pen-test web apps, and agents that scan GitHub repos and draft patches. The patching half is the one you can try.
Why it matters If you maintain code on GitHub, Defense Factory installs from OpenAI's plugin marketplace and hands you findings with patches for a person to approve. Point it at one repository you know well, and judge it by how many patches you merge untouched.
SRCOpenAI
Policy Safety2 min
Nonprofits sue five AI labs to freeze frontier models
Public Citizen and three other nonprofits asked a San Francisco court on Nov. 12 to bar Anthropic, Google, Meta, OpenAI and SpaceX's AI unit from building anything more capable than today's models. It is a long shot with a sharp second demand.
Why it matters Your API keys work on Monday. But the complaint builds its case from the labs' own warnings about their models, and it asks to ban AI-run AI research outright. Expect labs to weigh every published safety warning as possible evidence from now on.
SRCPlatkin LLP (complaint)

Prompt of the day

Find out if switching AI models saves you money

› I run an AI workload and need to decide whether to switch models. Act as a blunt cost analyst.

My current model: [name], at $[input price] and $[output price] per million tokens. Its price changes on [date] to $[new input] and $[new output] (write "no change" if none).
The candidate model: [name], at $[input price] and $[output price] per million tokens.
My workload: about [number] tasks a day. A typical task uses [input tokens] input tokens and [output tokens] output tokens on my current model.
The vendor says the candidate uses [X]% fewer tokens and finishes [Y] times faster. Treat both claims as unproven.

1. Compute my monthly cost on each model today and after the price change. Show the math.
2. Compute the token reduction the candidate needs to break even with my current model today.
3. Tell me what faster finishes are worth in my case. Ask me one question if you need to.
4. Design a one-week test on my own tasks: what to log per task (tokens in, tokens out, time to finish, pass or fail, minutes a human spent fixing it), how many tasks, and the rule that decides the winner.
5. End with one line: switch now, switch on [date], or stay.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

Google's Gemini 4 Hydrogen costs double Flash, for now

Jensen Huang quits Nvidia, agrees to testify to Congress

Anthropic's Claude Meetings joins your Google Meet calls

OpenAI's Daybreak adds agents that find and patch flaws

Nonprofits sue five AI labs to freeze frontier models