← The Weekly AI Recap

This week in AI: Nvidia loses Jensen Huang

November 15, 2026 · 4 min read

Good morning,

The company behind almost every AI chip lost its founder, Anthropic took back the top model spot, and AI agents got their own way to pay for things.

Nvidia's CEO quit, and AI startups may feel it first

A locomotive bearing the Nvidia logo keeps rolling with an empty driver's cab, while a line of small startup wagons behind it has one lime coupling unhooked.

Jensen Huang resigned as Nvidia's CEO on Thursday after 33 years. He left following reports of affairs, including one with a woman reportedly suspected of spying for China. CFO Colette Kress is interim CEO while the board searches for a permanent one.

The stock closed down 4.3% after falling as much as 16% during the day. Nvidia's board says it expects the product roadmap and customer commitments to continue without interruption, and earnings are still due on Tuesday.

The bigger effect may be on startup funding. Nvidia joined 105 funding rounds this year, often as the largest investor. Investors told TechCrunch that Huang made most of those calls and that Kress is "much more conservative."

Anthropic retakes the top spot, and Google opens up

The Claude mark sits on the tallest of three plain towers, while a neighbouring tower bearing the Gemini sparkle has its gate swung wide open.

Two labs released strong models within four days:

  • Claude Fable 5.2: Anthropic's new model is now first on Artificial Analysis's independent index. It is available to paid Claude plans, and the API price fell 20% to $8 per million input tokens.
  • Gemini 4 Carbon: Google's strongest model is no longer limited to cyber defenders. Paying developers and Ultra subscribers can use it at $2/$10, but the API price doubles on December 1.
  • Gemini 4 Hydrogen: a faster, cheaper model that edges past last month's Gemini 4 Argon. It is now in Google Search for everyone and has a free API tier.

The cheaper Fable price per token does not mean cheaper work. The model writes far more tests and checks, so its highest setting costs about $26 per index task in Artificial Analysis's tests, against $10 for Opus 5.5.

For hard problems, try Fable at its "high" setting. It scores almost as well as max at about a third of the cost.

AI agents can now pay for things

A small driverless cart carrying a shopping bag waits at a shop's checkout turnstile, which opens only after a lime fingerprint approval panel lights up.

Three launches this week let agents complete purchases, each with a human approval step:

  • ChatGPT Wallet: OpenAI's new wallet lets its dots agents buy things for US Pro subscribers. You approve each purchase with Face ID or a fingerprint, and the shop gets a one-time card number instead of your real one.
  • Replit Agent: now buys supplies for small businesses and runs their Google Ads. It stops at checkout until the owner approves the final total.
  • Brave Search: agents can now pay per search in stablecoins, at half a cent each, with no account or API key.

These are early, US-first launches, and some businesses still block agents outright. United Airlines now bans unauthorized ones from booking or checking in.

If you sell online, check that your bot protection doesn't block legitimate agent shoppers. OpenAI's partner playbook explains how to recognize them.

Quick hits 🗞️

  • Claude can now join your video calls. Claude Meetings takes notes and builds docs live in Google Meet, in beta for Team and Enterprise. It's switched on by default for Team plans, and Anthropic's own pages disagree on whether it joins calls with outside guests, so admins should check that setting before client calls. Source

  • One breach at an AI provider hit both Notion and Cursor. Attackers who got into Fireworks AI accessed chat transcripts from under 1% of Notion workspaces. Cursor says a key exposed at Fireworks was the starting point for the tampered Claude Code extension in its marketplace. If you pasted keys into either tool, rotate them. Source

  • Anthropic goes public this week. It plans to sell 825 million shares at $70 to $78 each under the ticker ANT. Pricing is set for Wednesday, with trading from Thursday. Source

  • Nonprofits sued five AI labs to freeze frontier models. Public Citizen and three other groups asked a San Francisco court to stop Anthropic, Google, Meta, OpenAI and SpaceX's AI unit from building anything more capable than today's models. Source

  • Google launched a cheap model for yes-or-no answers. The Gemini Decisions API classifies text, images, audio or PDFs for $0.10 per million input tokens and returns a confidence score with each answer. Independent tests show mixed results. Source

Model leaderboard 🏆

Via Artificial Analysis, which tests models independently.

  • Smartest:
    • Claude Fable 5.2 (65)
    • Gemini 4 Carbon (60), now open to paying developers
    • Claude Opus 5.5 (59)
  • Best value: GPT-6.1 Terra (51) at about $2 per million tokens, still the cheapest top-10 model per task.
  • Best open-weights: Moonshot's Kimi K3.1 Pro (46), free to download.
  • Moved this week: Fable 5.2 entered at #1, and Gemini 4 Hydrogen entered the top 10 at #7.

See you next week!

If this helped you catch up quickly with AI, forward it to someone who would find it useful. They can subscribe here.