Version v2.1.322 of Claude Code, a new release, is now out. This version lets users use fast mode on Google Cloud's Vertex. If Vertex rejects fast mode, the tool automatically switches to normal speed.
Anthropic has added new tools to its Claude AI: Whiteboards, a shared board for brainstorming, Sheets, a spreadsheet tool, and Data, for working with company information. All are available on paid plans. The main goal is to keep work inside Claude so teams can edit it together.
Whiteboards groups meeting notes by theme and turns plans into timelines with owners and dates.
Sheets builds tables with tabs, formulas, and drop-down lists from a text prompt, and tracks changes.
Data is now generally available, while certified metrics, data stories, and live sharing are in beta.
Results can be moved into documents, slides, or other tools like Miro and Airtable.
Anthropic introduced new Claude tools that let teams build spreadsheets, whiteboards, and data dashboards together. It also upgraded connectors so people can use apps like Adobe and Atlassian inside Claude without switching tabs.
Claude Sheets lets teams and Claude organize data together and create formulas.
Claude Whiteboards lets teams gather ideas in meetings, and Claude summarizes them.
Claude Data, formerly Dashboards, connects to tools like Snowflake and Salesforce to build dashboards.
Claude Meetings is coming soon and will take notes and update documents during calls.
Claude now lets you open tools like Asana, Canva, and Notion beside your chat, and it can make edits inside them. The feature is rolling out on web and desktop for all plans.
Claude can change a page or file you have open, and it only acts when you approve.
You can open a tool beside the chat or expand it to full screen.
You can pin your most-used tools to the sidebar for one-click access.
Developers can build MCP Apps, a shared way for tools to show their own interface in AI apps, and this is in public preview.
Anthropic has released v2.1.321, a new version of its AI coding tool Claude Code. It adds the new default model Claude Fable 5.2 and a feature to specify a settings file for remote control. Many security and error fixes are also included.
The new model Claude Fable 5.2 has been added and is now the default model. Its context window (the amount of text it can handle at once) is 1 million tokens.
Remote control commands can now specify a settings file. When the file changes, new sessions will not open until a restart.
Fixed an issue where authentication tokens could pass to other servers during MCP login, and an issue where read-only commands ran outside the sandbox.
Fixed several errors in notebook editing, file modification, and permission checks, where the wrong file could be changed or actions ran without asking.
Anthropic has upgraded its high-performance AI models, Claude Fable 5.2 and Mythos 5.2. The company says they are better at coding, reasoning (the process by which an AI works out an answer), and understanding images. Output is up to 30% faster than the previous version, and prices have dropped.
Output speed is up to 30% faster than the previous 5.1 version.
The price has fallen to $8 per million input tokens and $40 per million output tokens.
A fast mode (an optional setting that makes response generation up to 2.5 times faster) has also been added.
Fable is open to general users, while Mythos can be requested only for cybersecurity and life sciences research.
Anthropic reported on a test of whether its AI model Claude could design a circuit board from start to finish and build a working prototype. With the circuit board maker Diode Computers, Claude helped create three devices: a speakerphone, a voice assistant hub, and a handheld virtual pet. All three prototypes worked well.
Claude handled the specification, circuit diagrams, component placement, wiring, a 3D-printable case design, firmware coding, and the bill of materials.
Human engineers assembled the circuit boards and ran the firmware, following Claude's instructions.
Claude's spatial reasoning, which determines where parts go on the board and how copper traces connect them, improved greatly over previous models.
However, these are still prototypes, so long-term use and whether they pass commercial certification have not been confirmed.
Claude Code version 2.1.319 is out. It changes how auto mode works, adds a feature that runs MCP tasks in the background, and introduces several new settings. It also fixes some bugs.
When remote monitoring is turned off, auto mode no longer falls back to the default classifier without checking with the server. Instead, it asks the user or denies the action.
When an MCP tool call runs as a task, it is sent to the background right away, and the result is reported back to Claude when it finishes.
A new setting automatically denies tool calls that would trigger a permission prompt, so the work can continue.
The /tasks screen now lists finished subagents, so users can reopen them and send them messages.
Claude Enterprise has gained a security center, inference hooks, and centralized management features. Security administrators can review Claude settings at a glance, inspect inputs with their company's data security tools, and manage admin consoles in one place. The update targets businesses that want better visibility and control over security settings.
The security center compares a company's Claude settings with the settings recommended by Anthropic and shows the results.
Inference hooks send inputs and tool calls to the company's data security tools for inspection.
A single central management console lets administrators handle login and user management settings across multiple organizations at once.
Using the Admin API, administrators can log users out, directly manage IP allowlists, and block personal account use.
Claude has launched a beta personal finance feature that lets users connect bank and card accounts and ask questions about them by voice or text. Based on balances and transaction history, it organizes spending and also looks for unusual charges and fees. It cannot move money or make investments from the accounts.
Accounts are connected in read-only mode, so Claude cannot transfer money or manage the accounts.
Bank logins are handled directly through Plaid, a service that connects apps and banks. Claude never receives or sees passwords.
Claude does not sell loan, credit card, or deposit products directly, and there are no ads or recommendations.
For now, it is rolling out in the US to Pro, Max, and Ultra subscribers starting this week, in stages.
Anthropic has released version 2.1.318 of Claude Code, its AI coding tool. The update adds new features and fixes several bugs. Main changes include a way to check login expiration, connecting multiple AWS accounts, and better stability when running commands.
A command and warning message for checking when the login expires were added.
A setting was added that lets users borrow permissions from another AWS account.
Several problems were fixed, including logins being dropped and failures when saving settings files.
A problem on Mac where memory usage grew during many command runs, causing freezes, was also fixed.
New partner agents and products for security, development foundations, creative work, and productivity have been added to the Claude Marketplace. The marketplace now offers more than 25 partner products. Eligible enterprise customers can apply part of their committed spend with Anthropic toward these tools.
Security tools that find and fix vulnerabilities (security flaws) in code and systems have been added.
Tools that help development teams, such as deployment platforms and tools for tracing agent runs, have also been added.
Marketing and creative tools, including video and image generation and website management, are included.
Companies can split the same committed spend between Claude and partner products.
Anthropic has released v2.1.317, a new version of Claude Code, its AI coding tool. This version adds several new features along with security checks and bug fixes. It mainly improves enterprise management features and usage stability.
When creating a scheduled cloud task with the /schedule command, a warning appears if an unwanted external connection is attached.
Before the web or shell tools send the account's email address over the network, they ask for permission first.
In auto mode, a notification now appears even when a subagent (a helper AI that handles part of a task) gets no response to its tool call.
Fixed a problem where, when two permission prompts appeared at the same time, the second one did not show the file path or command.
A public beta of a "message threads" feature is now available on the Claude platform. With it, agents no longer have to resend the entire conversation history every time. As conversations get longer, the time Claude takes to start responding goes down.
With threads, only content added since the last turn is sent.
In requests of 500k to 1M tokens, the median time to first response dropped by about 35%, according to the company.
Threads expire five minutes after the last request, and after that the full conversation must be sent again.
Requests get smaller, but token usage stays the same because Claude reads the full conversation on every turn.
Anthropic has released v2.1.316, a new version of its AI coding tool Claude Code. The update adds a few configuration options and fixes errors that occurred with shell commands and in several connection situations. It includes both new features and fixes.
The CLAUDE_CODE_DISABLE_ULTRATHINK setting lets you turn off the ultrathink keyword (a command that tells the model to think especially deeply).
The fast mode of the Opus 4.8 model (a feature that returns responses more quickly) will be discontinued on November 30, 2026. After that, users need to switch to Opus 5.5 to keep the same speed improvement.
A problem where shell commands left running when Claude Code exited kept running on macOS and Linux has been fixed.
Several bugs have been fixed, including one where PowerShell hooks (tasks that run automatically in specific situations) got stuck on exit code 2.
Anthropic has released v2.1.314 of Claude Code, its AI coding tool. This version adds several new features and fixes many issues. It mainly focuses on file write restrictions and fixes for authentication and connection errors.
A new setting now prevents the AI from writing or modifying files outside the working folder.
When a conversation stays idle for a while, its content is summarized so it can be used in the next message.
A problem where the claude -p command quietly ended when it failed to start is fixed. It now shows an error and ends as a failure.
Fixed an error in sessions with many images and a problem where connections were blocked by security certificate errors.
IDC Research has connected its market analysis data so it can be used directly in Claude. Customers type in a question and receive answers based on IDC's forecasts and data. IDC also uses Claude every day across the entire company.
When customers ask Claude questions such as competitor analysis, vendor shortlists, or market size, they receive answers based on IDC materials.
Each answer shows the source, indicating which IDC document or dataset it came from.
IDC engineers rebuilt the internal document repository in 3 weeks instead of the expected 3 months.
The sales team has built a tool with Claude that reads CRM (customer relationship management) data to find deals that have stalled.
Anthropic has officially released three new features for the Claude Enterprise plan. They include reports showing how teams use Claude, surveys run inside the app, and shared budgets set up for each team. All three features are now available to everyone.
Smart reports look at parts of a team's conversations over a period the team chooses. They group the work into tasks and show the cost and results of each task, as well as the points where people often get stuck.
The reports also include recommendations for turning repeated workflows into skills (reusable ways of doing a task) or rules.
Admins can raise usage limits from the report. Team leads can also create reports for their own teams if they are given permission.
According to the article, the reports are meant to help with understanding adoption and planning investments. They will not be used for individual performance reviews or personnel decisions.
Anthropic has released artifacts (interactive tools created within a conversation) designed for legal teams to use right away. It has also opened a gallery of tools that help with tasks such as contract review, tracking regulatory changes, and handling incoming legal requests. Lawyers review the first draft Claude produces and make the final decisions.
It reviews a set of contracts, organizes the results into a table, and adds links to the original source of each item.
It checks whether citations in legal documents match the original sources and whether page numbers are correct, then flags any problems.
It finds clauses that differ from the company's standard contract terms and sends them to the approver.
Tools built by legal technology companies and law firms are also listed in the gallery.
Anthropic has released v2.1.312 of Claude Code, its AI coding tool. This update focuses on fixing errors and refining behavior rather than adding new features. Problems with session resuming, file reading, and command loading have been corrected.
Fixed an issue where large settings files such as CLAUDE.md were only partly read and then overwritten.
Fixed a problem in the VS Code extension where, after logging in, organization policies could not be fetched for the rest of the session.
Changed the usage limit notice from 'session limit' to '5-hour limit' for the 5-hour usage cap.
Prompt hooks and agent hooks are now limited to running up to 24 at the same time within one session.
Anthropic has expanded its partnership with the US General Services Administration (GSA). Government agencies can now use Claude products at a 40% discount. The offer applies to federal, state, local, tribal, and territorial governments.
The discount applies to Claude Enterprise, Claude for Government, and the Claude API. It applies whether agencies buy directly or through a cloud provider.
Agencies do not have to commit to a minimum spending amount and pay only for what they use.
The state of Utah gave Claude Code (an AI tool that writes and fixes code) to 34 developers and ran a four-week trial.
The company expects the discount to save the government about 1 billion dollars.
Anthropic has released a bundle of features called Claude for Financial Advisors. It pulls data from several financial firms into one place in Claude, helping advisors cut the time they spend on research and records before client meetings. The goal is to let them spend more time on direct work with clients.
Advisors spend only about one-sixth of their total work time meeting with clients, according to the company.
The new feature includes connectors that link Claude to customer management systems, brokerage firms, and asset management tools.
It also offers skills that help with tasks such as meeting preparation, portfolio analysis, and compliance checks.
Several financial companies, including Charles Schwab, BlackRock, and FactSet, are taking part as connection partners.
Anthropic has released a Claude plugin for nonprofit organizations. It is a bundle of features built together with about 30 nonprofits, and it is said to help with everything from fundraising to reporting while easing the workload of small teams. Anthropic also plans to expand the Claude Core program together with Taproot.
The plugin helps organizations use Claude in line with their own situations and rules.
It supports several stages, from fundraising research to reporting on results.
Claude Core is a program that places early-career fellows at nonprofits in the United States.
Through a partnership with Taproot, Anthropic plans to connect nonprofits with free consultants for short, project-based work.
A new version of Claude Code, v2.1.310, has been released. The main changes are the addition of a Monitor tool, several bug fixes, and improvements to cloud session management. Enterprise administrators can now set access permissions by role.
The Monitor tool can now be used in Bedrock, Vertex, and Foundry environments.
A problem where requests failed on some gateways has been fixed. When a request fails, the relevant value is removed and the request is retried.
A problem where tool calls were wrongly allowed on Windows without Git Bash installed has been fixed.
In cloud sessions, access can now be divided by enterprise user role.
Claude's regular chat and Cowork, its task mode, are now combined and officially available to Enterprise plan users. Both simple questions and long projects can be handled in the same chat window. The feature first came to Pro and Max plans in September.
Users no longer need to choose between chat and Cowork before starting.
Complex tasks keep running in the cloud managed by Anthropic, so work continues even after you close your laptop.
Admins can set access by role, and can manage usage with a new Efficient mode.
Existing security settings, such as SSO and the Compliance API, can still be used as before.
Anthropic says it has strengthened Claude's election-related safeguards ahead of the 2026 US midterm elections. It analyzed what people ask Claude about elections and politics, and added ways to connect users to trustworthy information.
About 0.51% of US Claude conversations were about elections or voting.
Questions about whom to vote for were very rare, at about 0.01%.
Users can get state-specific information on polling places and what to bring directly within the conversation.
On election night and for several days after, Claude will show the latest official vote counts.