What changed in AI tools this week: Gemini 4 Hydrogen, Claude Meetings and agent checkout

This week brought a cluster of updates that change how people use everyday AI tools rather than just which model tops a leaderboard. Google replaced its fast Gemini model, Anthropic pushed Claude further into meetings, spreadsheets and factory software, OpenAI gave its agents a way to pay, ElevenLabs entered the note-taking market, and the local AI engine llama.cpp reached version 1.0.

This roundup covers announcements made between November 10 and November 13, 2026. Each section says what changed, who can use it, and what it means if you are choosing or already paying for one of these tools. All facts come from the companies' own announcements or from named news reports, linked in each section. We have not tested these features ourselves; where a company makes a performance claim, we say whose claim it is.

Google: Gemini 4 Hydrogen becomes the everyday Gemini

What happened. On November 12, Google introduced Gemini 4 Hydrogen, the fast, lower-cost member of its Gemini 4 family. Google positions its bigger models, Gemini 4 Carbon and Argon, for long and difficult reasoning, and Hydrogen for quick execution in coding and agent tasks. According to Ars Technica, Hydrogen has taken the place of Gemini 3.8 Flash across Google's products.

Who gets it. Google says the rollout is global and includes the Gemini app, Google Search, Google AI Studio, the Gemini API and Gemini for business. Paid Google AI Pro and Ultra subscribers can also use it in Google Antigravity. Free Gemini app users reach it through automatic model selection, when the app decides the model fits the prompt.

Claims worth knowing. Google says Hydrogen completes tasks 2.5 times faster than 3.8 Flash, and reports an Artificial Analysis Intelligence Index score of 51 at the default thinking level and 54 at its new xhigh level. There are four thinking levels in total: low, medium, high and xhigh. For developers, Google lists $1.50 for every million input tokens and $7.50 for every million output tokens, with no surcharge for long prompts up to the 1 million-token context limit.

Search changes too. In a separate post, Google describes Hydrogen as the largest model upgrade Search has had. The company says Search should cope better with long questions that have several parts, run more parallel sub-queries to surface a wider set of links, and show more tables, images and structured answers.

What it means for you. If you build on the Gemini API's fast tier, this is the new default to evaluate, and Ars Technica notes the price matches 3.8 Flash. If you mainly use Gemini or Google Search as a consumer, you will likely notice the change without doing anything. Treat the benchmark numbers as Google's own, and compare outputs on your own prompts before switching production work.

Anthropic: Claude joins the meeting

What happened. Anthropic launched Claude Meetings in beta on November 12. Claude can join a Google Meet call as a visible participant, answer questions asked aloud or in the chat, take notes and action items, and build documents or whiteboards that colleagues edit during the call. After the meeting, a dedicated page in claude.ai holds the notes, the transcript and anything Claude created.

Who gets it. The beta is for Team and Enterprise plans. It is on by default for Team plans, and Enterprise admins switch it on. Google Meet is the only supported platform so far. Anthropic says organizations using zero data retention, HIPAA-ready configurations or customer-managed encryption keys cannot use it during the beta.

Privacy defaults. By default, Claude will not join calls that include people from outside the organization and will leave if an outside guest joins. It keeps a transcript, a summary and descriptions of the screen, but not audio. Anthropic advises checking that everyone on a call agrees to transcription before turning on auto-join.

Cost during the beta. Anthropic says notes, transcripts and in-call answers are free on Team and Enterprise plans, unless an answer uses a connected tool, which counts toward usage. Documents and whiteboards made during calls are free through February 10, 2027.

What it means for you. If your team already pays for a separate note-taker such as Granola, Fireflies.ai or Otter.ai, and you are on a Claude business plan, the beta is worth a trial on internal calls. Anthropic also says it is working with several of those companies so transcripts can flow into them, so the choice may not be either-or.

Anthropic: spreadsheets, whiteboards and full apps inside Claude

Two days earlier, on November 10, Anthropic shipped a set of workspace features that set the stage for Meetings.

• New artifact types. Claude Sheets lets a team and Claude build spreadsheets together, including formulas and multiple tabs. Claude Whiteboards is a shared board for laying out ideas. Claude Data, previously called Dashboards, connects to sources such as Snowflake and Salesforce to produce dashboards and data stories.
• Plugins with built-in skills. Anthropic lists plugins from Salesforce and Workday, both in beta, and from NetSuite, aimed at sales, finance and HR tasks.
• A large connector catalogue. According to Anthropic, Claude Marketplace now offers more than 9,000 connectors and plugins, some of them made by the community.

In a companion post, Anthropic explained that supported tools can now open beside the conversation or go full screen, with Claude making changes directly inside them after you approve. The first group includes Asana, Canva, Dropbox, monday.com, Vercel and Webflow, and Anthropic gives examples for Adobe, Atlassian's Confluence, Box, HubSpot and Notion. The feature is rolling out on the web and the desktop app on all plans. For developers, these views are built on MCP Apps, an open extension of the Model Context Protocol, so an app built this way is not tied to Claude alone.

What it means for you. The practical change is fewer copy-and-paste steps between Claude and the tools where work is stored. If your organization restricts connector tools, Anthropic says admins keep the same controls as before.

Anthropic: Claude for Manufacturing and self-hosted agents

On November 12 Anthropic also released Claude for Manufacturing, a plugin with 17 skills that reads data from Oracle NetSuite, Infor, Autodesk Fusion and Microsoft 365. The skills cover jobs such as drafting quotes, documenting failed inspections, ranking the likely causes of a breakdown and preparing supplier requests. Each is designed to end with a person's review. With Infor's own plugin, Claude can also place a purchase order with another supplier and reschedule production once a buyer approves. Anthropic says the plugin does not run physical equipment. It is available on Team and Enterprise plans.

For engineering teams, Anthropic announced self-hosted Claude Managed Agents in private beta. Managed Agents is Anthropic's set of APIs for building long-running autonomous agents. The self-hosted version lets a company run the whole service, including the harness, sandboxes, credentials and session history, in its own cloud or on its own servers, and send model requests to Anthropic directly, to a cloud provider such as Amazon Bedrock or Microsoft Foundry, or through its own gateway. Anthropic describes safeguards such as keeping credentials out of the sandbox, limiting network access to an allow-list of hosts, and exporting a record of every agent step over OpenTelemetry.

What it means for you. These are business offerings rather than consumer features. If you evaluate AI for regulated or security-sensitive work, the self-hosted option changes what is possible to approve; access currently goes through Anthropic's account team.

OpenAI: ChatGPT Wallet lets agents check out

OpenAI launched ChatGPT Wallet on November 13, as TechCrunch reported. It lets OpenAI's Dots agents complete a purchase for a user. At launch it is limited to some US ChatGPT Pro subscribers aged 18 or older with an eligible US-issued card. Before each checkout, the user reviews the items, the total and the card, then approves with Face ID or a fingerprint. OpenAI says the wallet pays with a payment token so the agent and the merchant never see the real card number, and that card rewards and dispute rights still apply.

The Verge points out a practical limit: some retailers restrict or block AI agents, so a Dot cannot shop everywhere. OpenAI says it wants to let users pre-approve purchases within spending limits, categories or time periods later on.

What it means for you. If you are outside the US or not on Pro, there is nothing to set up yet. If you are eligible, the per-purchase approval means the agent cannot spend without you, which is a reasonable starting point for trying agent shopping.

ElevenLabs: a meeting and dictation app for teams

According to TechCrunch, ElevenLabs released an experimental app called ElevenAI this week during its 11/11 Summit in New York. It combines meeting notes, search across past meetings, live help during a conversation, and dictation into any app with support for over 90 languages. Companies can connect ElevenAgent to take follow-up steps such as drafting an email from meeting notes. The app is in preview and requires signing up for early access.

What it means for you. Together with Claude Meetings, this shows meeting assistants moving from simple transcription toward taking actions. If you are about to renew a note-taker licence, it may be worth waiting to see how these previews mature.

llama.cpp 1.0: a steadier base for local AI

The ggml-org team released llama.cpp v1.0.0 and ggml v1.0.0 on November 13. The main change is a promise: from now on, breaking changes to the public C API will come with a major version bump, under published semantic versioning rules. The team says llama.cpp supports more than 150 model architecture variants and lists distribution through channels including Homebrew, Docker, Debian, conda-forge, MacPorts and Winget.

What it means for you. If you run models locally, you may not see new features from this release, but desktop apps that depend on llama.cpp should be able to update with fewer surprises.

A short checklist for the week

• Gemini API users: test Gemini 4 Hydrogen against your current fast-model prompts; Google lists the same price as 3.8 Flash.
• Claude Team or Enterprise admins: decide whether to allow Claude Meetings, and review the external-guest setting before enabling auto-join.
• Teams paying for a note-taker: compare Claude Meetings and ElevenAI previews with your current tool before your next renewal.
• ChatGPT Pro users in the US: if you try ChatGPT Wallet, start with small purchases from retailers that allow agents.
• Local AI users: watch for app updates built on llama.cpp 1.0.

All of these launches are recent and several are in beta or preview, so details such as eligibility and pricing may change. Check each company's linked announcement before making a buying decision.