What Anthropic shipped, translated.
Every Claude update we've covered, newest first. The homepage shows the last two months; this page keeps the rest.
Each card shows what was true when it shipped.
July 2026
Opus 5 — frontier-class work at the same price
Anthropic's new heavy-hitter landed July 24. On hard coding and knowledge work it comes close to Fable 5, the most capable model — at half Fable's price, and the same $5/$25 per million tokens as the Opus 4.8 it replaces. It's the default model on Claude Max and the strongest you can pick on Pro, and it checks its own work before handing it back. Everyday Free and Pro work still runs on Sonnet 5 — nothing to change there.
Cowork now runs in the cloud — start at your desk, finish from your phone
Cowork used to need the desktop app open. Since July 7 it's rolling out to the web (claude.ai) and the iOS and Android apps — beta, starting with Max, more plans to follow. The real change: your session runs on Anthropic's servers, so it keeps working after you close your laptop. Start a report or a spreadsheet job at your desk, then steer it or approve a decision from your phone while it runs. (Reading local files and controlling your computer still need the desktop app.) Anthropic doubled Cowork's usage limits through August 5, so there's room to test bigger tasks.
Building on Sonnet 5? Two things will break.
If your code sets temperature or top_p, those calls now error on Sonnet 5 — remove them before switching the model ID. And the new tokenizer produces up to about a third more tokens from the same text, so any token budgets calibrated on Sonnet 4.6 will undercount. The trade: promotional pricing ($2/$10 per million tokens) runs through August 31, which makes this a cheap window to migrate and test.
The Fable 5 saga: launched, suspended, back
Anthropic launched Fable 5 on June 9 — then had to suspend it three days later when the US government applied export controls to its newest models. The controls lifted June 30 and Fable 5 returned worldwide July 1. One thing to know: it's the first Claude model that costs extra credits to use — plan which tasks deserve it.
June 2026
Sonnet 5 is the new default
Open Claude on a Free or Pro plan and you're already working with Sonnet 5. It's faster and gets more done on everyday work — writing, analysis, research — without switching models. Nothing to set up; it's just what you're using now.
Claude is now inside Copilot
Claude models in Microsoft Foundry now include Sonnet 5 (added June 30). If you're using Opus 4.1 through Foundry, note it retires August 5 — Anthropic recommends moving to Opus 4.8. Claude still powers Researcher in Copilot and Excel Agent Mode.
Claude reads the messy stuff now
Claude reads messy spreadsheets, dense charts, screenshots, and scanned PDFs far more accurately now — useful when the clean source file doesn't exist and all you've got is a picture. Claude also holds long, multi-step work together across a session, so you don't have to re-explain the task halfway through.
Claude checks its own work now
Before handing back a finished task, Claude reviews its own work — checking the result against what you actually asked for, and catching its own mistakes first. You get a more finished draft instead of babysitting every step.
May 2026
Dynamic workflows — Claude runs a whole job in parallel
Claude breaks a huge coding job into many parallel sub-agents, then checks the work before reporting back. It plans the job first, runs the pieces at once in one session, and verifies the result. Research preview on Enterprise, Team, and Max.
Change Claude's instructions mid-task
The Messages API now accepts system entries inside the messages array — so you can update Claude's instructions partway through a task instead of starting the conversation over.
Agent View — one dashboard for every Claude Code session
Launched May 11. Run claude agents to get one screen for every Claude Code session — running, blocked, or done. Reply inline to unblock without leaving the dashboard. No more juggling tmux panes. Available on Pro, Max, Team, Enterprise, and Claude API plans.
Set a goal. Walk away.
Launched May 11. The new /goal command lets you define a completion condition and Claude keeps working across turns until it's met. Live overlay shows elapsed time, turn count, and token usage. Works in interactive, -p, and Remote Control modes.
Claude is in your Excel now
Claude for Excel, PowerPoint, and Word is now generally available for paid subscribers. Claude for Outlook is in public beta. Context carries between apps — what starts in a model ends in a deck without re-explaining.
Connectors went consumer
AllTrails, Instacart, Audible, Tripadvisor, Uber, and Spotify joined the connectors directory. These are personal-account connectors. Useful for life admin; check with your IT team before connecting anything to a work account.
Doubled rate limits
Pro, Max, Team, and Enterprise plans got their Claude Code five-hour limits doubled. Peak-hour throttling is gone for Pro and Max.
Claude went Wall Street
Launched May 5. Ten pre-built finance agent templates (pitchbooks, KYC, month-end close). Moody's MCP integration brings credit and risk data on 600M+ companies inside Claude. Plus eight new data connectors: Verisk, Third Bridge, Fiscal AI, D&B, Experian, GLG, Guidepoint, IBISWorld.
April 2026
Adobe, Blender, Ableton — Claude shows up where you create
Launched April 28. Claude now plugs into the tools creatives already use — Adobe, Blender, Autodesk Fusion, Ableton, and Affinity by Canva, plus four more. Brief Claude in chat, get help inside the app.
Custom charts and visualizations, right in the chat
Claude can now generate charts, diagrams, and data visualizations inline in its responses. No exports, no external tools — paste a dataset or describe what you want, and the visual renders in the chat. Great for quick analysis, meeting prep, and one-off dashboards.
See the guide →Claude Design — collaborate with Claude on slides, prototypes, and one-pagers
Launched April 17. Describe a slide deck, prototype, or one-pager — Claude builds it. Learns your brand from your codebase or uploaded files. Exports to PDF, PPTX, or Canva. Research preview for Pro, Max, Team, and Enterprise.
Claude Code desktop rebuilt around parallel sessions
New sidebar for every active and recent session. Integrated terminal, in-app editor, rebuilt diff viewer, drag-and-drop panes, side-chat shortcut (Cmd+;) for branching off a task without polluting the main thread.
Routines — Claude Code runs without you
Bundle a prompt, a repo, and connectors into one config that runs on schedule, on API call, or on a GitHub event. Runs on Claude Code's web infrastructure — laptop doesn't have to be open. Research preview for Pro, Max, Team, Enterprise.
xhigh effort, Auto mode for Max, /ultrareview
Opus 4.7 added an xhigh effort level, and Auto mode rolled out to Max. The /ultrareview command runs a dedicated review session that flags what a careful reviewer would catch.
Memory is now on for everyone — free included
Claude now remembers across chats for all users. For structured work that needs exact context control, Projects still wins — but casual chats carry forward automatically.
Cowork is fully live on Mac + Windows
Cowork hit general availability April 9 on the Claude Desktop app for macOS and Windows. Schedule recurring tasks. New Customize section combines Skills, Plugins, and Connectors. Enterprise gets Analytics API, usage dashboards, OpenTelemetry.
See the guide →Managed Agents — fully-managed agent harness
Launched April 8 in public beta. Anthropic-managed sandbox, built-in tools, server-sent event streaming. Use the managed-agents-2026-04-01 beta header. Paired with Routines, this is the full stack for hands-off agent work.
Skills go org-wide and cross-platform
Deploy Skills across your whole org. Partner Skills directory is live. The Agent Skills open standard means Skills work on other AI platforms, not just Claude.
See the guide →February 2026
No ads. Ever.
From Anthropic's Feb 4, 2026 post "Claude is a space to think": Claude will never show ads. No sponsored answers, no advertiser influence on what it tells you. When Claude recommends a tool, it's because it's the right one — not because someone paid.