What's new? — October 2026
The key recent Claude releases, with a link to the relevant module.
New
👥 AI and jobs at Anthropic (Oct. 16)
Anthropic publishes a study of its own teams: the share of juniors among technical hires falls from 27% to 9%, agents work in total about 6 times as many hours as their human handlers, and Claude spends 80% of interactive session time waiting for human input. Anthropic warns its practices don't transfer as-is, but the signals are worth pondering. → Module 15 ter
New
⚖️ AI Act: Brussels asks for answers (Oct. 14)
The Commission says the AI Office, which has been able to police model providers since August 2, has sent requests for information following incidents during testing this summer. According to Euronews, OpenAI and Anthropic are concerned; no penalty announced. → Module 13b
New
🔎 Code Review for Team and Enterprise (Oct. 15)
Review of code changes proposed on GitHub leaves research preview: according to Anthropic, most reviews are about 3 times faster (6 to 10 minutes instead of 20), with 3 depth levels per repository. An owner must turn it on. For your technical teams. → Module 06
New
⚡ Sonnet 5.5 (Sept. 28)
The everyday model changes: more than 30% faster than Sonnet 5 and up to 30% cheaper per task, at the same price ($2/$10), within 2 points of Opus 5.5 on GDPval-AA (real professional tasks), per Anthropic. Open to everyone, Free included. On a sensitive cybersecurity request, it visibly falls back to Sonnet 5. → Module 02
New
🪶 Haiku 5.5 (Oct. 7)
The small model takes a leap: about 75% cheaper than Haiku 4.5, 1M context, adjustable effort. Ideal for sorting, summarizing and classifying at scale. The same day, Sonnet 5.5 cache reads drop to $0.10 (about 20% less on agentic work), and Anthropic announces monthly API credits for Max and Team (rolling out gradually). → Pricing
New
🧩 Docs, Slides and Design for all (Oct. 8)
Out of beta: Claude Docs, Slides and Design are open to every plan, Free included (on Enterprise, on by default from October 15). New in beta: Dashboards, live dashboards wired to your data, and Motion, animations exported as MP4. → Module 05
New
📗 Claude in Google Workspace (Oct. 6)
The side panel comes to Google Docs, Sheets and Slides: formulas, pivot tables, charts, slides that follow your templates. Public beta on every paid plan, deployable by admins from the Google Admin console. → Module 09
New
📜 Usage policy (Oct. 8)
Effective November 12: for any use that affects a person's health, rights, finances or livelihood, a qualified human reviews Claude's recommendations and the person is told that AI was used. → Module 13b
New
🚧 Agents overstepping: the report (Oct. 9)
Anthropic documents actions its models took without being asked, mostly during tests (a form submitted, a block bypassed). Impact judged minimal; the lesson for your agents: written scope, limited access, human sign-off. → Module 15
🧠 Opus 5.5 (Sept. 22)
Fable 5.1-level performance on most tasks for 40% less than Opus 5 ($4/$20, cache reads $0.20), 30% faster, 1M context. Available on every paid plan and the new default Opus in Claude Code; five-hour limits raised on paid plans. Opus 5 becomes a legacy model. → Module 02
🧩 One Claude: Docs, Slides, Design (Sept. 16)
Cowork and chat merge: Claude decides on its own whether to answer, work on your files or produce a deliverable. Claude Docs, Claude Slides (export to PowerPoint or PDF) and Claude Design arrive in the conversation, in beta. Pro and Max first; on Enterprise, the admin decides. → Module 05
🧵 Projects with parallel threads (Sept. 17)
In Claude Code, a coordinator scopes the job, runs several threads in parallel (each on its own branch), tracks their progress and assembles the result, with shared memory and a shared library. Beta, Pro and Max, cloud sessions. → Module 06
🐢 "Pace the frontier" (Sept. 12)
In a long essay, Dario Amodei calls on the industry to deliberately slow capability gains so safety work can catch up, and pledges permanent access for independent evaluators. Opus 5.5 is the first model released since, tested before launch by METR. → Module 01
🦋 Fable 5.1 (Sep. 1)
The new quality ceiling: beats Fable 5 and Opus 5, same price ($10/$50) but cache reads 4× cheaper — up to −45% on agentic missions. Finer guardrails, included on Max. → Module 02
🔏 Text watermark (Sep.)
With Fable 5.1, every text Claude produces carries an invisible statistical watermark that cannot be turned off; generated files get C2PA metadata. It proves Claude was involved, not authorship. → Module 13b
🌐 Claude in Chrome GA (Aug. 26)
Every paid plan. Claude chains actions without per-step approval, a classifier validating each one; admin-approved domains, manual mode kept. → Module 07
🔗 Unified memory chat ↔ Cowork (Aug. 25)
One memory for chat and Cowork, as editable per-topic files. Sensitive topics stay excluded by default (explicit opt-in). → Module 13
🔄 The loop shift (August)
You no longer hunt for the right prompt: you write a loop with a stop criterion the machine verifies on its own, and Claude retries until it is met. New dedicated module: when to prompt, when to loop, and how to start without risk. → Module 03 ter
🌐 Cowork in Chrome (Aug. 12)
The extension's side panel becomes a Cowork session: Claude reads your tabs, clicks, fills out forms, and the task carries on to desktop or mobile. Max and Team, Pro rolling out. → Module 07
🎥 Record a Skill (July)
You do the task on screen while narrating it and Claude writes the Skill. Not a line to draft. Cowork in Claude for Mac, Pro, Max and Team plans. → Module 11
🧠 Opus 5 (Jul. 24)
Close to Fable 5's intelligence at half the price ($5/$25), 1M context, thinking on by default, verifies its own work. New default on Max. → Module 02
🚀 Cowork on web and mobile (Jul. 7)
Cloud sessions tied to the account: work continues with the machine off, synced across devices. → Module 05
🦋 Fable 5 restored (Jul. 1)
Global return after June's export suspension, with a new safety classifier. → Module 02
⚡ Sonnet 5 (Jun. 30)
Close to Opus 4.8 at 40% of the price — and the $2/$10 per MTok launch price is now permanent. Default model. → Pricing
🏷️ Claude Tag (Jun. 23)
Claude becomes a team member in Slack, delegable by @mention. Team/Enterprise beta. → Module 10
🔁 Claude Code: Routines & more
Cloud routines, /ultrareview, dynamic workflows, sandboxed built-in browser, automatic memory. → Module 06
📊 Microsoft 365 GA (May 7)
Word, Excel, and PowerPoint generally available, Outlook in beta, context shared across apps. → Module 09
🧠 Memory revamped (Jul. 10)
Categorized entries updated mid-conversation; Monthly Recap (Reflect) in beta. → Module 13
🛡️ Enterprise governance
Self-service Enterprise, RBAC, model entitlements, Compliance API, HIPAA-ready. → Module 13b