Claude features timeline: what shipped, where, and how to use it
Claude features ship almost every week, across half a dozen products. More than once I have found a feature days after release, after doing by hand exactly what it automates. This post is my fix: one page with a timeline of every notable feature on one side and the Claude products that use it on the other, the current version of each product, and for the features that matter most, the best use case and an example you can run today.
Last checked: October 17, 2026. A scheduled agent re-reads Anthropic's release notes every week, updates this post and emails me the changes. How that works is at the end.
TL;DR
- Six product surfaces share most features: the Claude apps (web, desktop, mobile), Cowork, Claude Code, the Claude Developer Platform (the API), Managed Agents, and the add-ins (Chrome, Excel, PowerPoint, Google Workspace, Slack).
- The current models are Claude Fable 5.1 (most capable), Claude Opus 5.5 (the default), Claude Sonnet 5.5 and Claude Haiku 5.5. Claude Code is at v2.1.302.
- The biggest shifts of the last year: agents that run without you (Routines, Cowork scheduled tasks, Managed Agents scheduled deployments), permission prompts replaced by auto mode, orchestration of many agents (workflows, background subagents, agent view), and Claude's output as live pages (artifacts, Docs, Slides, Design, Dashboards).
- If you adopt only five things: Skills, auto mode, Routines (or Cowork scheduled tasks),
effort, and artifacts.
The products and their current versions
| Product | What it is | Current version | Release notes |
|---|---|---|---|
| Claude apps (web, desktop, iOS, Android) | Chat, projects, memory, artifacts, Docs, Slides, Design, Dashboards | Opus 5.5 default; Docs, Slides and Design GA since Oct 8, 2026 | Claude release notes |
| Cowork | Claude doing multi-step work on your files and apps, from desktop, web or phone | GA on macOS and Windows (Apr 9), on web and mobile (Jul 7), from any chat (Sep 16) | Claude release notes |
| Claude Code (CLI, Desktop, web, VS Code, JetBrains) | The agentic coding tool | v2.1.302 | Changelog · What's new |
Claude Developer Platform (API, SDKs, ant CLI) | Messages API, tools, files, skills, batches | Models: claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5; Python SDK 1.x | API release notes |
| Claude Managed Agents | Anthropic-hosted agent loop and sandbox, on a schedule if you want | Public beta, header managed-agents-2026-04-01; dynamic workflows in beta since Oct 9 | API release notes |
| Claude Code Review | Multi-agent review on every pull request | GA since Oct 15, 2026 (Team and Enterprise) | Announcement |
| Claude in Chrome | Claude that browses and acts in your own Chrome | GA on all direct plans (since late June 2026) | Claude release notes |
| Claude for Excel, PowerPoint, Microsoft 365 | Office add-ins and connector | Add-ins GA; Microsoft 365 connector with write tools (Jul 7) | Claude release notes |
| Claude for Google Workspace | Docs, Sheets and Slides add-on and connectors | Add-on available; connectors in beta (Oct 6, 2026) | Product announcements |
| Claude Tag | Delegate tasks to Claude from Slack | Team and Enterprise (since Jun 23) | Claude release notes |
The timeline: feature ↔ products
Newest first. Products are where you can use the feature today, not only where it launched. The ★ rows have a use case and an example further down.
| Date | Feature | Products |
|---|---|---|
| 2026‑10‑15 | ★ Code Review GA: 3× faster, Quick/Standard/Deep per repo | Claude Code, GitHub |
| 2026‑10‑15 | Fast mode for Opus 5.5 leaves research preview | API, Claude Code |
| 2026‑10‑14 | OpenTelemetry export for API requests and agent sessions (beta) | API, Managed Agents |
| 2026‑10‑09 | ★ Dynamic workflows for Managed Agents (beta) | Managed Agents |
| 2026‑10‑08 | ★ Claude Dashboards (beta) and Claude Motion (beta); Docs, Slides, Design leave beta | Claude apps |
| 2026‑10‑07 | Claude Haiku 5.5: 1M context, cheapest and fastest model | All |
| 2026‑10‑07 | Max and Team plans include monthly API credits | API |
| 2026‑10‑06 | ★ Google Docs, Sheets and Slides add-on and connectors | Claude apps, Cowork |
| 2026‑10‑01 | ★ Mods: change Claude Code's behavior and UI with TypeScript | Claude Code (CLI, Desktop) |
| 2026‑09‑28 | Claude Sonnet 5.5 | All |
| 2026‑09‑25 | Plugin developer portal: submit plugins, see usage | Claude Code, Cowork |
| 2026‑09‑22 | Claude Opus 5.5, the new default model ($4/$20 per MTok) | All |
| 2026‑09‑16 | Cowork from any conversation; artifacts become Docs, Slides, Design | Claude apps, Cowork |
| 2026‑09‑15 | Salesforce in Claude (beta) | Claude apps |
| 2026‑09‑14 | Compaction on demand (beta) | API |
| 2026‑09‑10 | ★ Managed Agents auto permission policy | Managed Agents |
| 2026‑09‑08 | ★ claude plugin eval: test a plugin against a no-plugin baseline | Claude Code |
| 2026‑09‑03 | ant apply: agents and environments as code | API, Managed Agents |
| 2026‑09‑01 | Claude Fable 5.1; per-message effort (beta) | All; API |
| 2026‑08‑25 | Memory across chat and Cowork, with editable Topics | Claude apps, Cowork |
| 2026‑08‑19 | Agent Skills, Skills API and Files API leave beta; browser use toolset | API, all apps |
| 2026‑08‑18 | ★ /design: Claude Design artboards in the terminal | Claude Code |
| 2026‑08‑14 | ★ Auto mode becomes the default permission mode (Pro, Max, Team) | Claude Code |
| 2026‑08‑07 | Managed Agents session budgets, advisor model, skills from GitHub | Managed Agents |
| 2026‑08‑03 | Cross-session messaging; self-hosted cloud environments (beta) | Claude Code |
| 2026‑07‑24 | Claude Opus 5; mid-conversation tool changes (beta) | All; API |
| 2026‑07‑20 | Claude Security plugin; iOS Simulator pane in Desktop | Claude Code |
| 2026‑07‑13 | ★ Artifacts call your MCP connectors; public share links | Claude Code, Claude apps |
| 2026‑07‑07 | Cowork on web and mobile; Microsoft 365 write tools | Cowork, Claude apps |
| 2026‑07‑06 | In-app browser in Claude Code Desktop | Claude Code Desktop |
| 2026‑06‑30 | Claude Sonnet 5; Claude in Chrome GA; background subagents by default | All; Chrome; Claude Code |
| 2026‑06‑23 | ★ Claude Tag: delegate from Slack | Slack |
| 2026‑06‑15 | ★ Artifacts: live, shareable pages from a session (beta) | Claude Code |
| 2026‑06‑09 | Claude Fable 5; ★ Managed Agents scheduled deployments | All; Managed Agents |
| 2026‑05‑28 | Claude Opus 4.8; ★ Workflows (dozens to hundreds of subagents) | All; Claude Code |
| 2026‑05‑19 | Self-hosted sandboxes and MCP tunnels for Managed Agents | Managed Agents |
| 2026‑05‑18 | ★ /code-review and /usage breakdown | Claude Code |
| 2026‑05‑11 | ★ Agent view (claude agents); /goal | Claude Code |
| 2026‑05‑11 | Claude Platform on AWS | API |
| 2026‑05‑06 | Multiagent orchestration, ★ Outcomes, webhooks, Dreams | Managed Agents |
| 2026‑04‑23 | Memory for Managed Agents (beta) | Managed Agents |
| 2026‑04‑20 | ★ /ultrareview: a fleet of bug-hunting agents in the cloud | Claude Code |
| 2026‑04‑17 | Claude Design (Anthropic Labs) | Claude apps |
| 2026‑04‑16 | Claude Opus 4.7 with xhigh effort; task budgets (beta) | All; API |
| 2026‑04‑14 | ★ Routines: scheduled, GitHub-triggered or API-triggered cloud agents | Claude Code (web) |
| 2026‑04‑09 | Advisor tool (beta); Cowork GA | API; Cowork |
| 2026‑04‑08 | ★ Claude Managed Agents (public beta); ant CLI | API |
| 2026‑04‑06 | Ultraplan; ★ Monitor tool; self-paced /loop | Claude Code |
| 2026‑03‑30 | Computer use in the CLI (research preview) | Claude Code, Cowork |
| 2026‑03‑23 | ★ Auto mode (research preview) | Claude Code |
| 2026‑03‑13 | 1M context GA for Opus 4.6 and Sonnet 4.6 | API, Claude Code |
| 2026‑03‑12 | Interactive charts and diagrams inline in chat | Claude apps |
| 2026‑03‑02 | Chat-history memory for every plan, including Free | Claude apps |
| 2026‑02‑25 | ★ Cowork scheduled tasks; Customize section in Desktop | Cowork |
| 2026‑02‑24 | Cowork plugin marketplace | Cowork |
| 2026‑02‑19 | Automatic prompt caching | API |
| 2026‑02‑17 | Claude Sonnet 4.6 | All |
| 2026‑02‑05 | Claude Opus 4.6 with adaptive thinking; ★ effort GA; ★ compaction API (beta); Claude for PowerPoint | All; API; Office |
| 2026‑01‑29 | ★ Structured outputs GA | API |
| 2026‑01‑12 | Cowork research preview (macOS, Max) | Cowork |
| 2025‑12‑18 | ★ Agent Skills open standard and org-wide skills; ★ Claude in Chrome beta for paid plans | All; Chrome, Claude Code |
| 2025‑11‑24 | Claude Opus 4.5; Claude for Excel (beta); tool search; programmatic tool calling | All; Office; API |
| 2025‑11‑14 | Structured outputs (public beta) | API |
| 2025‑10‑23 | Memory for Max, then Pro | Claude apps |
| 2025‑10‑15 | Claude Haiku 4.5 | All |
Best use cases, with an example for each
Grouped by what they help you do, not by date. Each one says when it pays off and gives an example you can paste.
Teach Claude once: Skills, plugins and mods
Agent Skills (Dec 2025, GA on the API Aug 2026) · Claude apps, Cowork, Claude Code, API. A skill is a folder with a SKILL.md that Claude loads only when the task matches its description.
Best for: any procedure you explain more than twice: your release checklist, your brand voice, how your team writes migrations.
---
name: release-notes
description: Write release notes from merged PRs. Use when asked for release notes, a changelog entry or "what shipped".
---
1. List PRs merged since the last tag: `git log --merges $(git describe --tags --abbrev=0)..HEAD`.
2. Group them under Added, Changed, Fixed. Skip chores and dependency bumps.
3. One line per change, in the user's language, linking the PR.
Save it as .claude/skills/release-notes/SKILL.md in a repo (Claude Code) or upload the folder under Customize → Skills (Claude apps and Cowork). Then run /skill-doctor in Claude Code to see what each skill costs in context and how often it fires.
Plugins and claude plugin eval (Sep 2026) · Claude Code, Cowork. A plugin bundles skills, commands, hooks, agents, MCP servers and mods.
Best for: sharing a setup with a team, and proving a plugin actually helps before you ship it.
claude plugin eval init # drafts test cases and graders for your plugin
claude plugin eval # runs them and compares against a no-plugin baseline
Mods (Oct 1, 2026) · Claude Code CLI and Desktop. Small TypeScript functions that run before, after or instead of events such as tool calls, permission requests and screen drawing. Installed through plugins, hot-reloaded. Best for: house rules a prompt can't enforce: redacting secrets from tool output, blocking a command in one repo, adding a status panel.
Write a mod that redacts anything matching sk-[A-Za-z0-9-]{20,} from tool output
before it reaches the conversation, install it, and hot-reload it.
Mods are not sandboxed and run with Claude Code's access: install only ones you wrote or trust.
Stop babysitting permission prompts: auto mode and hooks
Auto mode (preview Mar 2026, default since Aug 14, 2026) · Claude Code, Managed Agents (auto permission policy). A classifier approves safe actions and blocks risky ones (destructive git, rm -rf on unresolved variables, transcript tampering).
Best for: long tasks where you would otherwise click "allow" fifty times, without --dangerously-skip-permissions.
Press Shift+Tab in Claude Code until the mode line reads auto. Add hard limits it can never override:
{
"permissions": {
"deny": ["Bash(git push --force:*)", "Bash(terraform apply:*)"]
}
}
Hooks · Claude Code, Agent SDK. Shell commands (or HTTP calls) that run on events. Since 2.1.295 a hook with "onFailure": "block" stops the action if the hook itself fails.
Best for: checks that must always run, like formatting every edited file.
{
"hooks": {
"PostToolUse": [
{
"matcher": "Edit|Write",
"hooks": [{ "type": "command", "command": "jq -r '.tool_input.file_path' | xargs npx prettier --write" }]
}
]
}
}
Let it run without you: Routines, Cowork tasks and scheduled deployments
These three overlap. Pick by where the work lives:
| You want | Use | Where |
|---|---|---|
| A coding task on a repo, on a schedule or on a GitHub event | Routines | Claude Code on the web |
| A recurring task on your files, email and apps | Cowork scheduled tasks | Claude apps (desktop, web, mobile) |
| An agent in your own product, on cron, billed to the API | Managed Agents scheduled deployments | Claude Developer Platform |
Routines (Apr 14, 2026) · Claude Code on the web. A saved prompt plus repo and environment, fired by a schedule, a GitHub event (pull request, release) or an HTTP call. Best for: nightly dependency bumps, triaging new issues, a weekly "what changed" digest. This post's update agent is one.
Every Monday at 8:00: read the release notes at platform.claude.com, support.claude.com
and code.claude.com/docs/en/changelog. Add every new feature since the "Last checked"
date to src/content/articles/claude-features-timeline, open a pull request, and email me
a summary.
Cowork scheduled tasks (Feb 25, 2026) · Cowork. Best for: non-code recurring work: a Monday inbox triage, a Friday spreadsheet report.
Every Friday at 4pm, pull this week's closed deals from the Q4 pipeline sheet,
chart them by region, and draft an email to me with the chart and three takeaways.
Managed Agents scheduled deployments (Jun 9, 2026) · Managed Agents. Each firing starts a fresh session with your agent and environment. Best for: the same thing as a Routine, but inside your own product and account.
curl -fsSL https://api.anthropic.com/v1/deployments \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "anthropic-beta: managed-agents-2026-04-01" \
-H "content-type: application/json" \
-d '{
"name": "Weekly release digest",
"agent": "'"$AGENT_ID"'",
"environment_id": "'"$ENVIRONMENT_ID"'",
"initial_events": [{"type": "user.message", "content": [{"type": "text", "text": "Summarize the Claude release notes from the last 7 days."}]}],
"schedule": {"type": "cron", "expression": "0 8 * * 1", "timezone": "America/Indiana/Indianapolis"}
}'
Many agents at once: workflows, agent view and background work
Workflows (May 28, 2026) · Claude Code; dynamic workflows in Managed Agents since Oct 9. Claude writes a script that fans work out to dozens or hundreds of subagents and collects the results. Best for: the same change across many files or repos, a review with several independent lenses, research across many sources.
Use a workflow: for every package in packages/*, upgrade to TypeScript 5.9, fix the
type errors, run that package's tests, and report which ones still fail.
Agent view, background subagents and /fork (May–Aug 2026) · Claude Code. claude agents shows every session: running, blocked on you, or done. Subagents run in the background by default, and /fork copies the conversation into a background session.
Best for: keeping three or four tasks moving without losing track of which one needs you.
claude agents
Monitor and /loop (Apr 2026) · Claude Code. Monitor streams background events (a log, a dev server) into the conversation; /loop repeats a prompt on an interval, or self-paces without one.
Best for: "watch the deploy and tell me when it's green", "re-run the flaky test until it fails, then debug it".
/loop 10m check the staging deploy; when it finishes, run the smoke tests and report
/goal (May 2026) · Claude Code. Keeps Claude working across turns until a condition holds.
/goal every test in tests/ passes and npm run lint is clean
Review before you merge: Code Review, /code-review and /ultrareview
Code Review (GA Oct 15, 2026) · GitHub, via Claude Code; Team and Enterprise. Reviews every pull request with several agents; admins pick Quick, Standard or Deep per repository. Best for: teams that want a consistent first pass on every PR.
/code-review (May 2026) runs locally on your diff as a background subagent. /ultrareview (Apr 2026) sends a fleet of bug-hunting agents to the cloud, and claude ultrareview runs it from CI.
Best for: /code-review before every push; /ultrareview before a risky merge.
claude ultrareview # in CI or a script: cloud review of the current branch
Claude's output as something people use: artifacts, Docs, Slides, Design, Dashboards
Artifacts (beta Jun 2026, MCP-connected Jul 2026) · Claude Code, Claude apps. A session's output becomes a live page on claude.ai that updates as the session works, can call each viewer's own MCP connectors, and can be shared publicly. Best for: status pages, reports for your team, small internal tools.
Publish an artifact that lists my open PRs across all repos, with CI status,
and refreshes when I open it.
Claude Docs, Slides and Design (GA Oct 8, 2026) and Claude Dashboards (beta) · Claude apps. Dashboards connect to BigQuery, Databricks, Snowflake or Salesforce. Claude Motion (beta, Team and Enterprise) turns a prompt into an animated explainer that stays editable as code. Best for: Docs for anything people comment on; Slides from a voice memo; Dashboards for a metrics page that stays live.
Build a dashboard from our Snowflake ORDERS table: weekly revenue, top 10 products,
and refund rate, with a filter by region.
/design (Aug 2026) · Claude Code. Drafts editable artboards for your UI and implements the one you pick.
Work where you already are: Chrome, Office, Google Workspace, Slack
Claude in Chrome (beta Dec 2025, GA Jun 2026) · Chrome, Claude Code. Claude browses with your own sign-ins. Best for: tasks on sites with no API: filling a vendor portal, comparing listings, testing your own app as a user.
Claude for Excel and PowerPoint, Microsoft 365, Google Workspace (Nov 2025 → Oct 2026). Best for: working inside the file instead of copying data into chat.
In this sheet, build a pivot of revenue by region and quarter, flag any quarter that
dropped more than 10%, and add a chart on a new tab.
Claude Tag (Jun 23, 2026) · Slack, Team and Enterprise. Best for: handing off "can someone look at this?" from a channel, with the thread as context.
For builders on the API
Claude Managed Agents (beta Apr 8, 2026) · Anthropic runs the agent loop and a sandbox per session; you create the agent once and start sessions. Best for: hosted agents with files, bash and MCP, without writing the loop or the infrastructure.
import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic();
const agent = await client.beta.agents.create({
name: "Release digest",
model: "claude-opus-5-5",
tools: [{ type: "agent_toolset_20260401", default_config: { enabled: true, permission_policy: { type: "auto" } } }],
});
const environment = await client.beta.environments.create({
name: "digest-env",
config: { type: "cloud", networking: { type: "limited", allow_package_managers: true } },
});
const session = await client.beta.sessions.create({
agent: { type: "agent", id: agent.id, version: agent.version },
environment_id: environment.id,
});
await client.beta.sessions.events.send(session.id, {
events: [{ type: "user.message", content: [{ type: "text", text: "Summarize this week's release notes." }] }],
});
Outcomes (May 2026) · Managed Agents. A separate grader iterates the agent against your rubric. Best for: one deliverable that has to meet a bar: a report, a PR, a dataset.
Effort and adaptive thinking (GA Feb 2026) · API, Claude Code (/effort). One knob for how hard Claude thinks. Opus 5.5 defaults to medium, so set it.
Best for: low for chat and subagents, high or xhigh for coding and long agent runs.
const response = await client.messages.create({
model: "claude-opus-5-5",
max_tokens: 16000,
thinking: { type: "adaptive" },
output_config: { effort: "high" },
messages: [{ role: "user", content: "Find the race condition in this file: ..." }],
});
Structured outputs (GA Jan 2026) · API. Responses guaranteed to match your JSON schema. Best for: extraction and anything a program parses.
const response = await client.messages.create({
model: "claude-opus-5-5",
max_tokens: 1024,
output_config: {
format: {
type: "json_schema",
schema: {
type: "object",
properties: { feature: { type: "string" }, date: { type: "string" }, products: { type: "array", items: { type: "string" } } },
required: ["feature", "date", "products"],
additionalProperties: false,
},
},
},
messages: [{ role: "user", content: "Extract the feature from this release note: ..." }],
});
Compaction (beta Feb 2026) · API, and automatic in Claude Code and the apps. Summarizes earlier context server-side when a conversation gets long. Best for: chat apps and agents that run for hours.
const response = await client.beta.messages.create({
betas: ["compact-2026-01-12"],
model: "claude-opus-5-5",
max_tokens: 16000,
messages,
context_management: { edits: [{ type: "compact_20260112" }] },
});
messages.push({ role: "assistant", content: response.content }); // keep the compaction block
Fast mode (GA for Opus 5.5 on Oct 15, 2026) · API, Claude Code. The same model at up to 2.5× the output speed, at a premium. Best for: interactive coding where waiting costs more than tokens. The beta header is no longer required for Opus 5.5, but it is still accepted, so this works on every fast-mode model.
const response = await client.beta.messages.create({
model: "claude-opus-5-5",
max_tokens: 4096,
speed: "fast",
betas: ["fast-mode-2026-02-01"],
messages: [{ role: "user", content: "Rename this function everywhere it is used: ..." }],
});
Advisor tool (beta Apr 2026) · API, Managed Agents. A cheaper model does the work and consults a stronger one when it is stuck. Best for: high-volume agents where Opus on every turn is too expensive.
{ "type": "advisor_20260301", "name": "advisor", "model": "claude-opus-5-5" }
Add that to tools on a request whose model is claude-sonnet-5-5.
How this post stays up to date
The reason I miss features is that release notes live in five places. So the checking is now a job for an agent, not for me.
- A scheduled task in Claude Code on the web (a Routine, from the table above) runs every Monday morning in a fresh cloud session with this repository checked out.
- It reads the official sources: the Claude release notes, the API release notes, the Claude Code changelog and weekly digest, and Anthropic's product announcements. Press and social posts only point it to things to verify there.
- It adds new rows to the timeline, updates the versions table and the "Last checked" date in both languages, runs the site's tests, and opens a pull request into
staging. - It emails me a summary: what is new, which products it touches, and a link to the pull request. If nothing changed, the email says so in one line.
The steps it follows live in the repository as a skill (.claude/skills/claude-features-update/SKILL.md), so the schedule only says "run the skill". If you want the same thing for your own stack, the cheapest version is a Cowork scheduled task with the prompt from the Routines section; the most controllable is a Managed Agents scheduled deployment.
Sources
- Claude release notes (Claude apps, Cowork, Chrome, Office)
- Claude Developer Platform release notes (API, Managed Agents, models)
- Claude Code changelog and What's new
- Product announcements
- Code Review is now generally available
- Customize Claude Code with mods