AI Tool Updates · November 2026

Claude Code 2.1.325 Claude Code now tells you how to continue after an organization policy blocks a request

When organization policy blocks a request, a line under the error now says to run /rewind and change the message, or run /clear.

What changes If a company policy stops your request, you no longer have to look up how to get unstuck.Source →

Claude Code 2.1.325 Claude Code subagents cut off at the output limit now retry with a shorter full answer

When a subagent hits the output token limit, it is now asked for the whole answer again, more briefly, instead of returning only the end.

What changes Delegating long research to subagents is less likely to hand back only the tail of the result.Source →

Codex Newly added rust-v0.169.0 · New Features Codex can fold finished work in its full-screen transcript

Turning on "Collapse turn details" in the full-screen transcript folds completed activity beneath each prompt while keeping answers visible. Tool output stays searchable.

What changes When you return to a long-running task, you can skim the answers without scrolling through dozens of lines of tool output, and search for a specific log when you need it.Source →

Codex Newly added rust-v0.169.0 · New Features Codex lets you add your own model provider from the terminal

You can add or edit custom model providers from the terminal. With custom gateway onboarding enabled, you can create a provider during sign-in and authenticate with a device code.

What changes If your company routes models through its own gateway, you can finish the connection at first sign-in instead of hand-editing config files.Source →

Codex rust-v0.169.0 · New Features Codex terminal workspace now supports multiple side-chat tabs

In the opt-in terminal workspace you can open several side-chat tabs and switch between them by keyboard or mouse. Saved local side chats can be reopened after restarting Codex.

What changes You can ask side questions without disturbing the main task, and because chats survive a restart it is easier to pick a line of questions back up.Source →

Codex rust-v0.169.0 · New Features Codex can search older tasks by name, prompt preview or working directory

The agent command center now lets you search older tasks by name, prompt preview, or working directory.

What changes When you run many tasks at once, it takes less time to find where yesterday's job lives.Source →

Codex rust-v0.169.0 · New Features Codex terminal can send images and pick a model in Cloud conversations

With the opt-in Cloud connection on macOS and Linux, you can send images, choose a model and reasoning effort, and archive or restore Cloud conversations from the terminal.

What changes You can paste a screenshot to explain a bug, or send easy jobs to a lighter model, without leaving the terminal.Source →

Codex rust-v0.169.0 · Bug Fixes Codex keeps the current session if handing an approved plan to a fresh one fails

Choosing "clear context and implement now" hands the full approved plan to a fresh session, and the current session stays open if the handoff fails.

What changes If the handoff fails after you polished a long plan, you do not lose that plan or the conversation.Source →

Gemini Gemini now works on Apple CarPlay

The Gemini app can now be used on Apple CarPlay.

What changes On a drive, you can call up Gemini from the car's screen to talk through a schedule or an idea hands-free, instead of reaching for your phone.Source →

Claude Code 2.1.322 Claude Code fast mode now works on Vertex

Fast mode is available on Vertex in the global, us and eu regions, falling back to standard speed if Vertex refuses it.

What changes Teams running Claude Code through Vertex get faster responses. If Vertex refuses, work continues at standard speed, so turning it on is low-risk.Source →

Claude Code 2.1.321 claude remote-control accepts a settings file

Passing --settings <file> to claude remote-control applies that settings file to every session it starts. If the file changes, no new session starts until you restart claude remote-control.

What changes You no longer need to match permissions and settings for each remote session by hand. Remember to restart after editing the file.Source →

Claude Code 2.1.321 Claude Code installs 2 to 4 seconds faster

claude install and the install scripts finish 2 to 4 seconds sooner.

What changes A small gain that only people who frequently install on fresh machines or containers will notice.Source →

Gemini Gemini Intelligence features reach Pixel 10 and a Motorola phone

Google expanded Gemini Intelligence features and supported devices. The November Pixel Drop brings the latest features to the Pixel 10 line, and the motorola signature 27 opens for U.S. pre-order on November 19.

What changes Pixel 10 owners can use features such as Rambler, which turns spoken thoughts into organized text. If you often jot notes on the move, install the update first.Source →

Gemini Gemini 4 Carbon arrives in the Gemini app

Google made Gemini 4 Carbon, its strongest model, available to developers, enterprises, scientists and consumers. Consumers get it in the Gemini app with a Google AI Ultra subscription.

What changes If you subscribe to AI Ultra, you can hand long multi-step tasks to Carbon. API pricing goes up on November 30.Source →

Gemini Use Adobe AI in Gemini and edit PDFs from Drive

Google and Adobe linked their tools. You can edit PDFs from Drive and use Adobe AI inside Gemini.

What changes It removes the step of opening Acrobat separately to fix a PDF, which suits anyone who often edits contracts or reports received by email.Source →

Claude Code 2.1.321 Claude Code now defaults to Fable 5.2 as its Fable model

Claude Code 2.1.321 uses Fable 5.2 as the default Fable model, with a 1M-token context window at $8 per million input tokens and $40 per million output tokens.

What changes For work that feeds a long codebase into one session, you get the newest model without picking it yourself, and input is billed at a 20% lower rate than the prior version.Source →

Claude Code 2.1.321 Claude Code's /model picker no longer renumbers its rows

In the /model picker, "More models" is now an unnumbered row at the bottom. Older models open under it without renumbering the rest.

What changes If you pick models by number key, expanding the older ones no longer shifts the numbers, so you are less likely to select the wrong model.Source →

Claude Code 2.1.319 Claude Code: /claude-api latency-optimize finds slow spots in your API code

A new command finds where your Claude API code makes people wait and tries faster settings against your evals.

What changes If users complain your Claude-powered app is slow, this shortens the time spent deciding where to start.Source →

Claude Code 2.1.319 Claude Code: environment variable to stop feature-flag fetching without disabling telemetry

Setting CLAUDE_CODE_DISABLE_FEATURE_FLAGS stops feature-flag fetching without turning off telemetry; every flag then uses its built-in default.

What changes Useful if server-side experiments make behavior shift from day to day and you want to pin defaults without touching other data-collection settings.Source →

Claude Code 2.1.319 Claude Code adds a setting to deny instead of waiting on permission prompts

With the permissions.denyInsteadOfPrompting setting on, a tool call that would wait on a permission prompt is denied and the turn continues. It applies to terminal sessions, subagents and teammates.

What changes It keeps an unattended job from sitting overnight on a single prompt. The trade-off is that a denied call does not run, so you need to check the result.Source →

Claude Code 2.1.319 Claude Code moves long MCP tasks to the background in the terminal

When an MCP server runs a tool call as a task, Claude Code moves it to the background right away and tells Claude the result when it finishes. The CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND environment variable makes it wait instead.

What changes You lose less time waiting on slow MCP jobs before you can do anything else.Source →

Claude Code 2.1.319 Claude Code /usage now shows spend for today, this week and this month

For developers behind a Claude apps gateway with no spend limit, /usage now shows spend for today, this week and this month, provided the gateway runs this version or later.

What changes Even on an account with no limit, one command shows what you have spent this month, so you can get a sense of it before the bill arrives.Source →

Codex rust-v0.167.0 · Bug Fixes Codex closes network approval prompts once requests end

Network approval prompts close when their requests end, and concurrent requests to the same endpoint now prompt one at a time.

What changes Fewer stacked approval prompts to sort through, so long agent sessions need less hand-holding.Source →

Codex rust-v0.167.0 · New Features Codex can now use your computer's system proxy settings

Codex 0.167.0 lets you opt in, from /experimental, to the macOS or Windows system proxy settings.

What changes On machines where internet access must go through a proxy, such as a company network, Codex can follow the computer's existing setting instead of needing separate configuration.Source →

Codex rust-v0.167.0 · New Features Codex: expand patch details while an approval prompt is open

While an approval prompt is open, you can expand patch details directly in the transcript.

What changes You no longer have to dismiss the approval prompt to see what you are approving, which saves steps when reviewing long changes.Source →

Codex rust-v0.167.0 · New Features Codex: sign in with an API key or cloud provider

API key and cloud provider options now sit under "Sign in another way." Enabling gcp_setup_wizard sets up Google Cloud Vertex AI.

What changes If you want to run Codex through your company's Google Cloud account, you can start setup from the sign-in screen instead of hunting through docs.Source →

Codex rust-v0.167.0 · New Features Codex now previews attached images while you type

In supported terminals, a local image you attach now shows a preview as you type. You can also click image paths in the transcript to open the file.

What changes When you paste a screenshot to explain an error, you can confirm it is the right file before you send it.Source →

Codex rust-v0.167.0 · Bug Fixes Codex: Ctrl+L clears the screen but earlier messages stay on Ctrl+T

Ctrl+L now clears only the terminal display, and you can bring back earlier messages with Ctrl+T.

What changes During a long session you can tidy a cluttered screen without worrying about losing earlier instructions.Source →

Codex rust-v0.167.0 · New Features Codex marks unread results in one agent command center

The codex agents --no-daemon command opens an agent command center, and both it and the session picker now mark results you have not read yet.

What changes When you run several tasks at once, you spend less time hunting for which one finished with a result you have not looked at.Source →

Codex rust-v0.167.0 · New Features Codex keeps your draft when you switch backends

You can switch between connected backends from the terminal session picker, and the draft you were typing is kept when you switch.

What changes You no longer have to delete or re-paste a long prompt just to change backends.Source →

Codex rust-v0.167.0 · Bug Fixes Codex keeps saved chats visible after you switch model providers

Saved chats stay visible after you switch model providers. If a chat cannot resume, you can read its transcript and prepare an editable prompt to continue in a new chat.

What changes Past work no longer seems to vanish after you change providers, and a chat that cannot resume can still be carried into a new one.Source →

Gemini Gemini video generation now runs up to 15 seconds, and costs 30% less

Gemini Omni 1.2 Flash generates and edits video up to 15 seconds long, at a 30% lower price.

What changes You can make a product teaser or ad spot of up to 15 seconds in one go, so there is less stitching of short clips. Because you can change the background or lighting by chatting, it suits producing several draft versions.Source →

Claude app Claude can now connect to your bank and card accounts to answer questions about your money

You can link bank, credit card, loan and investment accounts to Claude to see balances and spending on one page and ask questions about them. It is a beta, available only in the US on Pro, Max and Ultra plans.

What changes You would use it at month end to ask where your money went. Since it hands account data to an AI, check which accounts it can see before linking, and note it is not available outside the US yet.Source →

Claude Code 2.1.318 Claude Code now tells you when your claude.ai sign-in expires

claude auth status now shows when the stored claude.ai sign-in ends, and claude -p JSON results include the expiry date. Plain-text claude -p output prints a one-line warning starting 10 days before expiry.

What changes If you run Claude Code on a schedule, you can now avoid a sign-in that silently lapses and stops your jobs. The warning appears ten days ahead, which is when to log in again.Source →

Copilot Microsoft 365 Copilot extensibility Microsoft 365 Copilot's Researcher can now pull in HubSpot data

The Researcher agent now supports the HubSpot connector and can use information from HubSpot in its answers. It was not integrated before.

What changes When preparing a research brief before a sales meeting, you can get a draft that already reflects the customer history stored in HubSpot. The trade-off is that HubSpot data now mixes into the report, so check it before sharing.Source →

Copilot Teams Teams' Workflows app gets AI automation templates

The Workflows app in Teams for Windows and Mac gets AI-powered templates for automating complex tasks. An admin has to turn the feature on.

What changes Reports and checks you repeat in the same order can be handed to a template. If your company admin has not enabled it, you will not see it.Source →

Copilot Microsoft 365 Copilot Edge's new tab page gets a Copilot search box and suggested actions

On Microsoft Edge's refreshed new tab page, search, chat and web browsing run from one box, alongside Copilot-suggested actions and curated work content. Users can also see personalized news and activities.

What changes The first screen you see on opening the browser changes from a search page to a Copilot chat box, so if your company uses Edge it helps to tell your team in advance why the start page looks different.Source →

Copilot Microsoft 365 Copilot Copilot Cowork can now use your local Edge browser, including sites you are signed in to

When a task needs the web, Copilot Cowork can use Microsoft Edge on your device to open pages, gather information and fill in forms. It works on sites you are already signed in to, with your same access.

What changes Repetitive entry in a company portal that requires login can be handed off. The trade-off is that it acts with your access, so watch its progress in the Cowork conversation.Source →

Copilot Outlook Outlook's Calendar Copilot pane now lets you pick which account it uses

If you have more than one Copilot-enabled account in new Outlook, the Copilot pane in Calendar now shows an account dropdown. Before, the account followed the wider Outlook context.

What changes For people who use a work account and another organization's account, it reduces the mistake of asking about the wrong account's calendar.Source →

Claude Code 2.1.317 Claude Code warns when a scheduled task picks up connectors you didn't ask for

/schedule now warns when a new cloud scheduled task gets connectors you did not ask for, and tells you how to remove them.

What changes You can see at creation time whether an unattended task could reach your mail or document tools, which saves auditing permissions later.Source →

Claude Code 2.1.317 Claude Code now asks before a tool call sends your account email to the network

A check now asks before a web or shell tool call sends your account email address to the network.

What changes You can stop your email address from leaking through scripts or outside tools, at the cost of a few more approval prompts.Source →

Claude Code 2.1.316 Claude Code /code-review now shows a findings card

For /code-review started from the Code tab in Claude Desktop or on the web, cloud and remote-control sessions now show a findings card instead of a plain-text list.

What changes Findings are shown item by item, so it takes less time to spot what to fix in a long list.Source →

Claude Code 2.1.316 Claude Code can turn off the ultrathink keyword

In Claude Code 2.1.316, setting the CLAUDE_CODE_DISABLE_ULTRATHINK environment variable turns the ultrathink keyword off.

What changes A single setting stops the word ultrathink from slipping into a prompt and making responses slower or using more tokens.Source →

Claude Code 2.1.316 Claude Code will remove fast mode for Opus 4.8 on November 30

Fast mode for Opus 4.8 has been deprecated since October 30 and will be removed on November 30. After removal, fast-mode requests return an error rather than falling back to standard Opus 4.8.

What changes Anyone running Opus 4.8 with fast mode on could see jobs fail with errors after November 30. Moving to Opus 5.5 keeps the speedup.Source →

Claude Code 2.1.316 Claude Code lets you set the effort level subagents run at

The CLAUDE_CODE_SUBAGENT_EFFORT environment variable sets the effort level subagents use when nothing else names one. Before, they inherited the session's level.

What changes You can run the main task at high effort and keep side subagents light, which can cut waiting time and usage in long sessions that spawn many subagents.Source →