Claude Tag is not free. Anthropic does not publish a flat Claude Tag add-on fee, but a small business still needs a qualifying Claude Team or Enterprise organization, a funded usage balance for channel work, and a Slack workspace. Direct messages follow a different billing path: they normally use the sender's Claude seat rather than the organization's Claude Tag balance.
For a small team, the sensible buying decision is not “What is the per-user add-on?” It is “Can one shared, channel-scoped AI teammate remove enough recurring coordination work to justify Claude seats, metered usage, Slack costs, and the permission overhead?” Claude Tag can fit operations-heavy teams that already work in Slack and can give it narrow, well-maintained access. Skip it if your work lives mainly outside Slack, you need predictable flat pricing, or nobody will own permissions, memory, routines, and spend controls.
Pricing, plan, model, availability, migration, credit, permission, and product claims were rechecked against Anthropic and Slack documentation on October 2, 2026. Claude Tag remains a public beta, and no hands-on product test was performed.
Claude Tag's real cost is a stack: qualifying Claude access, metered channel usage, and whatever Slack plan the business already uses.
Claude Tag pricing: what a small business actually pays
Claude Tag has no published flat add-on fee, but “no add-on fee” is not the same as free. The cost stack has four parts.
| Cost layer | Verified October 2, 2026 | What it means for a small business |
|---|---|---|
| Claude organization | Team: Standard seats are $25 monthly or $20/month on annual billing; Premium seats are $125 monthly or $100/month annually. Team requires at least two members. Enterprise pricing varies. | A two-Standard-seat Team organization creates a $50 monthly floor, or $40/month on annual billing, before Tag usage and tax. |
| Claude Tag channel usage | Consumption-based; channel work draws from the organization's funded usage balance. Team must fund the balance before work runs. | Set an organization cap and per-channel caps. Anthropic publishes no reliable per-task estimate. |
| Claude direct messages | Normally billed to the sender's connected Claude seat and subject to that seat's limits. Some DMs from unconnected members can temporarily draw from the organization balance. | DMs can create a separate seat-planning problem; the organization Tag cap does not normally govern connected-user DMs. |
| Slack | Free: $0. Pro: $8.75/user monthly or $7.25/user/month annually. Business+: $18/user monthly or $15/user/month annually. Enterprise+: contact sales. | If Slack is already paid for, treat it as an existing platform cost. If not, include it in total cost of ownership. |
Anthropic's Team plan documentation says Slack users do not each need a Claude seat to use Claude Tag in channels. A small company could therefore start with the two-seat Team minimum while allowing more Slack members to use selected channels. That economy changes if employees need personal Claude DMs, Claude Code, Cowork, or other seat-based features.
The launch credit should no longer be included in a new budget. Anthropic offered one shared credit of $2,500 for Team organizations and $25,000 for Enterprise organizations, but its launch promotion expired at 11:59 p.m. Pacific on October 1, 2026. It also required a GitHub connection for organizations qualifying on or after September 2. Any old balance should be confirmed in the admin console; do not treat expired promotional credit as recurring value.
Which models and token prices affect Claude Tag usage?
Claude Tag lets a team choose among the Opus and Sonnet models allowed by its organization, set a channel default, switch within a thread, and see the active model in each reply footer. The exact list is organization-specific, so ask @Claude what models can I use here? instead of assuming every published model is enabled.
Anthropic does not publish a separate Claude Tag token-price table or a dependable per-task price. The public API list rates below are the current official model-price reference, while the Claude Tag List price analytics column is the authoritative way to observe actual channel cost during a pilot.
| Relevant published model rate | Input | Output | Important note |
|---|---|---|---|
| Claude Sonnet 5.5 | $2 per million tokens | $10 per million tokens | Lower-cost current Sonnet reference. |
| Claude Sonnet 5 | $2 per million tokens | $10 per million tokens | Anthropic made its introductory rate permanent. |
| Claude Opus 5.5 | $4 per million tokens | $20 per million tokens | Higher-cost reasoning option. |
| Claude Opus 5 or 4.8 | $5 per million tokens | $25 per million tokens | Older current Opus list rate. |
| Opus 5.5 fast mode | $8 per million tokens | $40 per million tokens | Research preview; higher speed and cost. |
| Opus 5 or 4.8 fast mode | $10 per million tokens | $50 per million tokens | Research preview; availability depends on the organization. |
These rates come from Anthropic's current model pricing and Claude Tag model guidance. Tool calls, long context, caching, connected-service charges, and negotiated terms can change the final bill. A long thread that repeatedly loads channel history, files, and tool results will not cost the same as a short summary.
Dated Claude Tag cost scenarios for small teams
The scenarios below are budget ceilings, not usage forecasts. They use US list prices checked October 2, 2026, exclude tax, ignore temporary Slack promotions, and assume Standard Claude Team seats billed annually.
| Scenario | Claude base | Slack base | Tag usage cap | Maximum planned monthly stack |
|---|---|---|---|---|
| Lean pilot: 5 Slack users, Slack Free, 2 Claude Team seats | $40 | $0 | $75 | $115 |
| Paid Slack pilot: 5 users on Slack Pro, 2 Claude Team seats | $40 | $36.25 | $100 | $176.25 |
| Operations rollout: 5 users on Slack Business+, 2 Claude Team seats | $40 | $75 | $250 | $365 |
The usage cap is a hard planning control, not an included allowance. Anthropic says Team organizations need a funded balance before channel work runs. Invoiced Enterprise organizations can accrue usage without a preset upper bound unless an Owner adds a cap. New channels inherit a default per-channel limit, and admins receive alerts at 75% and 95% of a limit.
When a cap is reached, Claude declines work rather than quietly completing only part of it. Admin analytics split channel spend into engaged, proactive, scheduled, and monitoring activity. Reading channels and deciding whether to respond is currently not billed; a working session that Claude starts proactively is billed.
What Claude Tag does differently from a personal AI assistant
Claude Tag is a shared identity attached to selected Slack channels, not merely one employee's chatbot pasted into Slack. In a channel, everyone works with the same Claude, can steer the same thread, and sees the result. The assistant can use organization-provisioned connections, remember channel context, run asynchronous work in an ephemeral sandbox, schedule routines, watch channels, and follow up when a thread stalls.
This shared operating model matters more than the chat interface:
- Channel work belongs to the organization. The agent uses channel-scoped access bundles and organization service credentials.
- Direct messages belong to the individual. They use the person's Claude account, connectors, permissions, and seat limits.
- Ambient behavior is on by default per channel. Claude may answer an untagged top-level message when it judges a response useful; an @mention guarantees pickup.
- Routines keep working. Scheduled digests, channel watching, pull-request subscriptions, and unresolved-thread follow-up use the channel's permissions.
- Slack is the launch boundary. Claude Tag works in Slack today. Anthropic says Microsoft Teams support is coming, but that is not current availability.
Because the product is in public beta, treat these capabilities as current documented behavior, not an uptime guarantee or permanent contract. Features, limits, controls, and defaults may change before general availability.
Permissions, memory, and attribution risks
Claude Tag's main risk is not that it can chat; it is that a shared agent can remember context and act through connected tools. The safest design gives it a dedicated service identity, read-only access by default, and separate scopes for sensitive channels.
Permissions and memory inherit across scopes. A private channel can add narrower tools, but it can still read workspace notes created from public channels.
| Area | Documented behavior | Small-business risk | Safer control |
|---|---|---|---|
| Organization and workspace access | Organization credentials can apply everywhere; workspace access applies to public channels and inherits organization access. | One broad credential can expose more systems than a pilot requires. | Use a dedicated service account and a pilot-specific access bundle. |
| Private-channel access | A private channel can add credentials and repositories on top of inherited access. | “Private” does not automatically mean isolated from all workspace context. | Audit inherited access before adding sensitive tools. |
| Channel memory | Anyone in the channel can read, add, correct, or remove channel memory. | Stale instructions or casual statements can become operating assumptions. | Review memory weekly and store long playbooks in controlled documents instead. |
| Workspace memory | Claude may save workspace notes from public channels and read them in public and private channels. | A public-channel convention can influence sensitive-channel work. | Owners should inspect workspace memory before a private pilot. |
| Ambient replies | Claude reads channel messages and can respond without being mentioned; automatic replies are on by default per channel. | Noise, surprise cost, or action from loosely phrased requests. | Turn Respond automatically off during the pilot. |
| Connected tools | Channel sessions use admin-provisioned connections; a user may approve a personal connector for their own request. | Mixed identities can confuse who authorized an action. | Keep personal connectors out of the first pilot and inspect audit trails. |
| Guests and shared channels | Guest channels default to blocked. Slack Connect channels are not supported. | A permissive guest setting may expose full configured access. | Keep guest response mode on Restrict; exclude external channels. |
| Attribution | Slack posts come from Claude; GitHub commits and pull requests come from the Claude GitHub App and link to the originating thread. Other tools use the provisioned service account. Enterprise can attribute channel spend to Slack users through an API. | Service-account activity can hide the human requester if logs are not reviewed together. | Retain Slack threads, service logs, and Claude Tag Activity records. |
Only a Claude organization Primary Owner or Owner can perform the initial setup. Role-restricted Claude Tag access is an Enterprise feature. Owners can also inspect scheduled and one-time tasks plus network calls on the Activity page, and can view, edit, or delete stored memory.
Best workflows for a small business
Claude Tag fits work that is already discussed in Slack, has clear source systems, and produces a reviewable artifact. It is less suited to irreversible actions, poorly documented processes, or work that depends on one person's private context.
Operations coordination
Use a private operations channel to summarize unresolved work, compare open threads with a task tracker, and draft a daily exception report. Let Claude read; require a person to approve assignments, purchasing, refunds, or schedule changes.
Sales follow-up
Have Claude collect unresolved deal questions from a channel, retrieve approved CRM context through a least-privilege connection, and draft follow-up notes. Do not let it change deal stages or send customer messages during the pilot.
Support triage
In an internal intake channel, Claude can group duplicate reports, identify known-answer requests, and flag items that need a human. Keep customer-facing replies and account changes behind review.
Reporting
Schedule a weekly summary from approved dashboards and project systems. Require every figure to link back to its source, and sample-check the report before expanding the audience.
Unresolved-thread follow-up
Ask Claude to post a daily list of open decisions, missing owners, and stalled approvals, skipping items marked with an agreed reaction. This is often a better first use than broad “watch everything” automation because the output is visible, reversible, and easy to score.
How to run a safe private-channel pilot
A useful pilot tests one bounded workflow for two weeks. It should measure value and risk before Claude gets broader access or automatic-reply privileges.
Start in one private channel, use read-only tools, require approval, and decide from observed cost and correction data.
- Create one private pilot channel. Include an Owner, the workflow owner, and two or three real users.
- Turn automatic replies off. Require an explicit
@Claudemention for every task. - Inspect inherited memory and access. Remove stale workspace notes and unnecessary organization-wide credentials.
- Connect one read-only system. Use a dedicated account; do not reuse an administrator's credential.
- Set budgets. Add a modest organization cap and a tighter channel cap. Record the starting balance.
- Choose the cheaper allowed model first. Use Sonnet for routine classification and summaries; test Opus only on tasks that actually need it.
- Run 20 to 30 representative tasks. Include normal, ambiguous, stale-data, and permission-denied cases.
- Review memory and Activity weekly. Confirm stored assumptions, routines, network calls, and source attribution.
- Decide from stop/go thresholds. Do not expand because the demo looked clever.
Practical go criteria include at least 80% of outputs usable with minor edits, zero unauthorized writes, source links on every factual report, and a stable list-price cost per accepted output. Stop or redesign if more than 20% of outputs need substantial correction, users cannot tell which identity performed an action, the channel exceeds its budget without a clear value driver, or the workflow repeatedly needs permissions broader than the business is willing to grant.
Migrating from the older Claude in Slack app
Claude Tag replaces the older per-user Claude in Slack experience in place. The Slack app and @Claude handle remain, but channel identity, memory, sessions, and ownership change. Anthropic says no data migrates.
The old app acted through each user's linked Claude account. Claude Tag uses one organization-managed agent identity in channels, persistent thread sessions, channel and workspace memory, and routines. Existing personal connectors and repository grants do not automatically become channel access; an Owner must provision new connections and GitHub access.
Enterprise organizations can phase the switch by scope using New, Legacy, or inherited version settings. Anthropic says the Legacy version is being deprecated, but directs customers to their account team for the cutover date. Team organizations use a single Enable Claude Tag switch where no scope routes to the older app.
Claude Tag alternatives for small teams
Choose an alternative based on the operating model you need, not on a generic AI feature checklist.
- Slackbot and Slack AI: better when the primary need is search, summaries, recaps, and workflow generation inside Slack without a separate Claude organization and agent identity.
- Claude Team without Claude Tag: better for personal research, writing, Cowork, and Claude Code when shared channel ownership is unnecessary.
- ChatGPT Business or Microsoft 365 Copilot: better when the team's source of truth is a broader office suite rather than Slack channels.
- Zapier, Make, or n8n: better for deterministic event-driven automation across many apps, especially when predictable triggers and explicit step-by-step logic matter more than conversational collaboration.
- A help desk or CRM's built-in AI: better when the work must stay inside a system with native permissions, audit history, and records.
Fit or skip: the small-business verdict
Claude Tag is worth piloting when Slack is already the operating room for the business, a recurring workflow is visible in channels, and someone can own access, memory, routines, and budgets. The shared identity and channel context can reduce copy-paste work and make asynchronous follow-up visible to the whole team.
Skip it for now if the business wants a flat, predictable add-on price; relies heavily on Slack Connect; has no Owner available to administer access; cannot create dedicated service accounts; or expects beta software to run consequential workflows without review. The winning use case is not “AI everywhere.” It is one measurable channel workflow with narrow permissions and a budget ceiling.
FAQ
Is Claude Tag free with Claude Team?
No. Anthropic publishes no separate per-seat Tag add-on, but channel work is consumption-based and requires a funded organization usage balance. Team seats and Slack may also cost money.
Does every Slack user need a Claude Team seat?
No for channel use. Anthropic says Slack users can work with Claude in channels without each holding a Claude seat. Team itself requires at least two members. Connected-user DMs use the individual's Claude seat.
How much does one Claude Tag task cost?
Anthropic publishes no dependable per-task figure. Cost varies with the selected model, input and output length, channel context, tools, and task duration. Use a capped pilot and monitor the per-channel List price data.
Can Claude Tag read every Slack channel?
No. An Owner selects where Claude Tag operates. It reads messages in channels it belongs to, subject to guest, shared-channel, scope, and response settings.
Can Claude Tag run scheduled and proactive work?
Yes, in the current public beta. It can run routines, watch channels, follow a pull request, and respond ambiently. Scheduled and proactive working sessions can consume organization usage.
Does Claude Tag work in Microsoft Teams?
Not yet. Anthropic says Teams support is coming, but Claude Tag works in Slack at the time of verification.
What happened to the Claude Tag launch credit?
The published promotion expired at 11:59 p.m. Pacific on October 1, 2026. New buyers should budget without it unless their admin console shows a valid remaining credit under separate terms.
Was this article based on hands-on testing?
No. This is a vendor-document-verified editorial analysis of current Anthropic and Slack documentation. Anthropic's capability descriptions are interested-party evidence, and beta behavior may change.