Anthropic ships Claude across 19 surfaces and documents each one on its own. Nowhere published is the comparison. This is that comparison, rebuilt daily.
Parity is an independent project, not affiliated with, endorsed by, or sponsored by Anthropic. Claude and Anthropic are trademarks of Anthropic PBC, referred to here only to describe what the tools do.
Pick as many as apply. These are the capability rows themselves, so every answer below traces back to a source.
Nothing selected yet. Pick a capability above and the answer appears here.
5 of 20 capabilities separate these two on confirmed facts. The remaining 15 they handle the same way, so they are not part of the decision and are not listed.
Nothing. Everything the other does, this does too.
Marked rows are the ones your selected surfaces disagree on, which is where the decision sits. Select a cell to see the detail and the plan gating behind it.
| Capability | Cowork | Claude Code | Agent SDK | Managed Agents |
|---|---|---|---|---|
Direct local file access Read and write files on the user machine without upload. | Yes | Yes | No | |
Live artifacts Artifacts that update in place during a session. | No | No | No | |
Chat memory Carries context across ordinary conversations. | No | |||
Persistent versioned memory API-addressable memory store with version history. | No | No | ||
Subagents Ephemeral isolated-context agents with restricted tools. | ||||
Agent teams Multiple full instances sharing a task list and messaging each other. | No | |||
Lifecycle hooks Intercept tool calls at lifecycle events. | ||||
Plugin marketplaces Browse and install plugins with versioning and updates. | ||||
Agent Skills SKILL.md packages, auto-invoked or called by name. | Yes | Yes | ||
Scheduled tasks Recurring runs on a cron-like schedule. | ||||
Outbound webhooks Signed subscribable event delivery. | No | No | ||
Sandboxed execution OS-level isolation of tool and code execution. | ||||
Computer use Direct screen control, clicking and typing. | No | No | ||
Browser control Navigate, click and fill forms in a real browser. | No | No | ||
Code execution Run shell commands and code. | Yes | Yes | Yes | Yes |
Voice mode Spoken conversation, not just dictation. | No | No | ||
MCP servers Connect external tool servers. | Yes | |||
Session resume and fork Continue or branch a previous session. | Unknown | |||
Batch processing Async bulk request processing at a discount. | No | No | No | No |
Zero Data Retention eligible Can be covered by ZDR arrangements. | Unknown | Yes |
The part official documentation cannot tell you, because each surface is documented on its own.
Chat memory works in web, desktop and mobile chat but does not transfer into Cowork. Inside Cowork, memory exists only at the Cowork Projects level. Any scheduled Cowork task that needs to remember yesterday requires an external state store.
Live artifacts, local MCP server plugins and direct filesystem access are desktop only. Cloud sessions reach files solely through a brokered connection to a running desktop app, so a closed laptop removes the capability entirely rather than degrading it.
Install a plugin in ordinary web chat or the desktop chat tab and its subagent and hook components appear inactive. The plugin looks installed while the orchestration parts do nothing.
Computer use in Cowork, cross-device task assignment and the mobile location tool are Pro and Max only, and explicitly unavailable to Team and Enterprise. This runs opposite to the usual gating direction and is easy to get wrong when advising on plan choice.
Multiple full Claude Code instances sharing a task list and messaging each other exist only in the CLI. SDK documentation states plainly that this is not configurable through SDK options. Managed Agents multiagent orchestration is a separate construct, not the same feature exposed elsewhere.
The CLI installs from marketplaces with versioning and auto-updates. The Agent SDK loads plugins from a local filesystem path only, so you vendor them yourself. Managed Agents has no plugin concept at all.
Sessions are stateful and stored server-side by design, so Managed Agents is explicitly ineligible. Running the CLI or SDK against the standard API keeps your existing ZDR or HIPAA arrangements intact. For regulated work this single fact can decide the architecture before anything else does.
Claude Projects sync across devices and are shareable on Team and Enterprise. Cowork Projects are desktop only, stored locally, do not sync, cannot be shared even on Team or Enterprise, and are not usable in Claude Code.
They run in the cloud, so they work with connectors and account-saved files only. Anything needing local files executes only if the desktop app happens to be reachable at that moment.
Signed subscribable webhooks covering session, agent, deployment, environment and memory events, with retries and deduplication. The CLI and SDK have no outbound equivalent. CLI Channels is inbound and a research preview, so it is not a substitute.
CLI and SDK auto memory is machine local, filesystem based, tied to a repo path, with no API, no versioning and no sync. The Managed Agents Memory Store is an API resource, mountable by multiple agents and sessions, with immutable version history.
The CLI has native per-command OS sandboxing on macOS, Linux and WSL2, with no native Windows support. The SDK needs a separate package. Managed Agents isolates at the whole-sandbox level instead, either an Anthropic-managed cloud sandbox or a self-hosted worker polling a queue.
The CLI and interactive SDK use can ride a Claude subscription, but products built on the Agent SDK are explicitly barred from offering Claude login or subscription rate limits to their own end users and must use an API key. Managed Agents has no subscription path at all.
Available in web, desktop and mobile chat. Cowork and Claude Code offer dictation only.
Written daily by an automated watcher across every published changelog. Summaries only, each linking to the page it came from. Nothing is reproduced verbatim.
The Enterprise plan overview lists two minimums depending on how you buy: 20 seats self-serve, 50 seats sales-assisted. The separate seat management article states a flat 20 seat minimum and draws no distinction between the two purchase paths. Both pages are live and neither references the other. Anyone sizing an Enterprise purchase from the seat article alone will under-budget by 30 seats on the sales-assisted path. This settles a question that had been open on the Parity site since the matrix was compiled.
Plan advice only. No capability changes on any surface. Resolves open question 5, which was previously recorded as having no public source for the figure 50.
The matrix recorded Cowork as off by default on Enterprise and enabled per group. The live admin article says Cowork is on by default for both Team and Enterprise, and that owners can disable it manually. What is off by default on Enterprise is a separate toggle, Run Cowork in the cloud, which an owner turns on and then grants to a group via custom roles. Team has that toggle on by default and no group controls at all, so Cowork on Team stays all or nothing. The practical effect is that Enterprise users get Cowork immediately but not cloud sessions.
Narrows a gap that was recorded as wider than it is. The Enterprise restriction applies to cloud execution, not to Cowork as a product. Scheduled cloud tasks are the thing gated on Enterprise, which is exactly the mode this watch runs in.
The Office add-in family has grown from Excel and PowerPoint to four apps, adding Word and Outlook, plus a cross-app mode that lets Claude read from one M365 app and write to another in a single conversation. Cross-app mode is on by default for Pro and Max and off by default for Team and Enterprise, with an org-level toggle under Organization settings, Office agents. Skills enabled in Claude settings now apply inside all four add-ins during a cross-app task, so a skill enforcing modelling conventions runs in Excel while a template skill runs in PowerPoint. Claude can only touch files already open, and cannot create, open, close or switch files.
Opens two gaps. First, cross-app mode does not work through Bedrock, Vertex, Foundry or an LLM gateway, so it is direct-account only, which widens the cloud integration parity gap. Second, the existing exclusion from Enterprise audit logs, the Compliance API and data exports now spans four apps rather than two, and the add-ins do not inherit custom data retention settings. Also worth noting the tracked surface list of 19 no longer covers everything: Word and Outlook are untracked.
The matrix recorded Claude Code as included per seat across Team and Enterprise. The live article adds a carve-out for older Enterprise contracts, where Claude Code is reachable only through specific seat types: Chat plus Claude Code seats billed on usage, or Premium seats billed per seat. Newer self-serve Enterprise plans do bundle it with the standard seat, matching what the matrix said. This matters when advising an org on an existing contract rather than a new one.
Plan gating only. The capability set of Claude Code is unchanged on every surface.
The matrix recorded Claude Design as web only. The current getting started article says it is available on web and desktop, and can be opened from the Claude desktop sidebar. Combined with yesterday's reclassification from research preview to beta, the product is further along than the matrix reflected. Verified 2026-08-17: the getting started article states web and desktop, but the admin guide for Team and Enterprise states Claude Design is available only through the web interface at claude.ai/design. The two articles contradict each other and neither references the other, so desktop availability is not settled.
Minor correction. No other surface gains or loses anything.
The getting started article says Claude Design can be used on the web at claude.ai/design or from the sidebar in Claude Desktop, and lists availability as web and desktop. The admin guide for Team and Enterprise restricts it to the web interface alone. Both pages are live, neither references the other, and no third article breaks the tie. Parity records desktop availability as unsettled rather than picking a side. Both articles agree the product is beta on Pro, Max, Team and Enterprise, and default off for Enterprise.
No capability moves on another surface. It does mean a Claude Design workflow cannot be assumed to work from the desktop app without testing it first.
The Enterprise seat article documents the two purchase paths separately. Self-serve plans cannot reduce seats mid-term and cancel from Organization settings. Sales-assisted plans go through an account manager for seat changes and through Sales to cancel. This article states a minimum of 20 seats.
Affects plan advice only. No capability on any surface changes as a result.
The Claude Design admin guide lists the product as beta across all four paid tiers, and off by default for Enterprise. Parity previously recorded it as a research preview, so this is a lifecycle step rather than a capability change. The guide does not state the Team default, which stands as previously recorded.
A lifecycle step rather than a capability change. Nothing gains or loses a feature on any other surface.
Anthropic has added invisible watermarking to Claude text output to meet EU AI Act requirements. Stated as having no effect on output quality, speed or cost. Relevant to anyone republishing or reprocessing model output downstream.
Applies across all surfaces rather than to any one of them.
Listed rather than guessed at. A visible unknown is worth more than a confident guess.