Nvidia names an interim CEO as Claude starts joining video calls
Nvidia’s board confirmed Jensen Huang’s resignation as president, CEO and director and named CFO Colette Kress interim CEO. Google released Gemini 4 Hydrogen for everyday coding and tool use. Anthropic began rolling out Claude as a visible participant in Google Meet.
01
Industry
2 stories
2026-11-12NVIDIA
Jensen Huang resigns and CFO Colette Kress becomes interim Nvidia CEO
Nvidia’s board said on November 12 that Jensen Huang had resigned as president, CEO and director, effective immediately. Executive vice president and CFO Colette Kress will keep those titles while serving as interim CEO. The board will consider internal and external candidates for a permanent successor.
Kress has led Nvidia’s finance organization since joining in 2013. The board also appointed Harvey C. Jones as its first chair. Nvidia said it still plans to report fiscal third-quarter 2027 results on November 17.
2026-11-12House Ways and Means Committee Democrats
Two Democratic lawmakers seek records on Nvidia chip export approvals
U.S. Senator Ron Wyden and Representative Richard Neal wrote to Commerce Secretary Howard Lutnick on November 12 to announce an inquiry into how the Trump administration approved exports of Nvidia’s advanced AI chips to China and handled related levies.
The lawmakers requested preservation of three sets of records: the Section 232 investigation into semiconductor imports and resulting actions, Nvidia export-license deliberations and approvals, and communications involving Nvidia, Jensen Huang or their representatives. The letter sets out the scope of the inquiry. The inquiry has yet to issue findings.
02
Models and APIs
3 stories
2026-11-12Google DeepMind
Gemini 4 Hydrogen launches for everyday coding and agent workflows
Google DeepMind released Gemini 4 Hydrogen for everyday coding, file retrieval and multi-step tool use. It is available through developer surfaces including the Gemini API and AI Studio, and is available in the Gemini app and Google Search. API pricing is $1.50 per million input tokens and $7.50 per million output tokens across the one-million-token context window.
The model offers low, medium, high and xhigh thinking levels and defaults to high. In Google’s own comparison, Hydrogen at its default high setting completed benchmark tasks 2.5 times as fast as Gemini 3.8 Flash.
2026-11-12Thinking Machines Lab
Thinking Machines publishes Inkling 2 Small weights under Apache 2.0
Thinking Machines Lab released the Inkling 2 Small weights on Hugging Face under the Apache 2.0 license. At launch, the team offered access through Tinker first and said it intended to release the weights after external testing.
The weights can now be downloaded from the Hugging Face model card. It includes an example with image and text inputs alongside benchmark results.
2026-11-12World Labs
World Labs opens its Atlas world model to the public
World Labs opened Atlas to the public after an early-access period. Its web apps and developer API can generate or reconstruct scenes, with output including keyframes, video clips and 3D Gaussian splats.
Workflows can start with one or more images to generate new viewpoints or reconstruct a space. A 3D Gaussian splat can be taken into downstream tools and viewed from different angles.
03
Products and developer tools
5 stories
2026-11-12Anthropic
Claude Meetings enters beta as a visible Google Meet participant
Anthropic is rolling out Claude Meetings in beta on Team and Enterprise, starting with Google Meet. Claude joins as a visible participant. People can ask questions aloud or in chat and ask it to work on documents and whiteboards during the call.
Claude asks for permission before using a participant’s connected tools. Answers posted in the meeting chat are visible to everyone on the call. Notes, action items, documents and whiteboards are available on the meeting page afterward. Team admins can turn the feature off, while Enterprise admins have to enable it.
2026-11-12Anthropic
Claude makes Opus 5.5 the default on four paid plans
Opus 5.5 is now the default model in Claude for Pro, Max, Ultra and Team. Anthropic recommends it for research, analysis, writing, and preparing documents, slides or spreadsheets that require judgment.
Opus 5.5 defaults to medium effort, which users can change in the model menu. Token-based billing is $4 per million input tokens and $20 per million output tokens. Anthropic says higher-priced models draw down usage limits faster on those subscriptions.
2026-11-12Google
Gemini-powered eyewear opens for preorders ahead of December 10 launch
Preorders have opened for Gemini-powered eyewear from Google, Samsung, Warby Parker and Gentle Monster. The glasses go on sale online and in stores on December 10. Warby Parker frames start at $395 and Gentle Monster frames at $449.
The glasses offer hands-free translation and visual questions, and require a compatible phone and an eligible 18+ Google account from the United States or South Korea. Select partner stores offer demos.
2026-11-12Replit
Replit Agent can prepare purchases and Google Ads campaigns
Replit Agent can now select products and fill a cart on a merchant’s website, then ask the user to approve the total. Stripe Link and Visa support the payment flow. Purchasing starts with Core and Pro users in the United States and is limited to one-time purchases.
The ads feature is available to Pro users worldwide. Agent can use a site or app to draft ad copy, generate images and recommend a daily budget. The user reviews the campaign before launching it across Google surfaces including Search and YouTube.
2026-11-12Figma
Figma opens its 3D transforms beta to Full seat users
Figma’s 3D transforms are now in open beta for all Full seat users. Text, images, vectors and frames can be tilted or rotated. A perspective effect on a parent frame puts multiple layers in the same spatial composition.
The original artwork stays editable. Figma’s brand team combined the effect with shaders and keyframed animations in Figma Motion for the launch video, testing how layers moved before settling on a composition.
04
Research and field tests
4 stories
2026-11-12Epoch AI
Epoch AI estimates compute at Huawei’s Horinger campus from satellite imagery
Using satellite imagery and public sources, Epoch AI estimates that the operational AI compute at Huawei’s Horinger campus is equivalent to about 81,000 Nvidia H100s. Its 90% confidence interval spans 36,000 to 141,000 H100 equivalents. Using the same measure, the researchers put the campus at about 7% of Tennessee’s Colossus 2.
The estimate draws on visible data-center and cooling infrastructure and converts chips to a common performance unit. Campus-wide capacity may be spread over multiple clusters. The compute available to train one model depends on how those clusters are connected.
2026-11-12OpenRouter
OpenRouter details the cost of translating its site into seven languages
OpenRouter described how it maintains its localized site. Each day it identifies changed English strings, translates them, runs code checks and verifies them with a model. Translations are stored with the code, and a person merges the daily pull request. Reader and reviewer reports can become glossary rules, style notes or regression cases.
At the current 16,918 English strings, the team estimates about $460 in translation and verification calls to retranslate the entire site into seven languages. Engineering work, additional review calls and human merging are outside that figure.
2026-11-12OpenRouter
OpenRouter tests how six models respond to repeated tool-call limits
OpenRouter ran the same failing-tool task three times on each of six models with repeat detection on and off. With detection off, four of 18 runs retried until the 40-step cap. With it on, all six models were blocked on their third identical call, and 16 of 18 runs made only two actual tool calls.
The write-up also shows how to treat variants of the same record identifier as one input and set step and cost caps for a run. OpenRouter recommends an API key credit limit to cap combined spending across concurrent sessions.
2026-11-12Anthropic
Anthropic interviews eight startups about using Claude Tag in Slack
Anthropic published a Claude Tag guide based on interviews with eight startups. It describes how teams use Slack to triage alerts, answer internal questions and hand off customer accounts. At Campfire, an engineer tagged Claude about undelivered agent replies. A pull request appeared 21 minutes later and was merged that afternoon.
Claude Tag is in public beta on Team and Enterprise. Connecting it requires a Claude organization owner and a Slack workspace administrator. Team plans also require usage credits.