Claude Tag (2026): What @Claude in Slack Costs, Who Gets It, and Who Pays

The short answer

Claude Tag is @Claude inside Slack, and as of October 25, 2026 it's only on Claude Team and Enterprise, in beta. There's no separate subscription and no published per-request price: work done in a channel is billed to the company by usage, and a direct message bills to the person's own seat. At Anthropic's published model rates, the everyday requests cost cents. The thing to watch is that seat limits don't cap channel spend. The only cap is the one an Owner sets.

Most of what's written about Claude Tag is the demo: tag it in a messy thread and it sorts out who owes what. That part works. What the launch material doesn't put on one page is the billing. So below I've costed three normal requests at the published rates, worked out a month for a 20-person company, and listed the three settings I'd set before anyone types @Claude.

What Claude Tag does

Claude in Slack switched over to Claude Tag on August 3, 2026. From Anthropic's product page and help center, read today:

• Tag it into any thread. Ask it to catch you up, run the numbers, file a ticket or turn a chaotic chat into action items. Everyone in the channel sees the work and anyone can steer it.
• Standing instructions. It can watch a channel, post a weekly digest, flag anything urgent, or tag you when a thread goes quiet, without being asked each time.
• Its own identity. It acts under its own account in your tools, not a borrowed login, and every credential use is logged. Commits and pull requests it opens show the Claude GitHub App as the author.
• Memory per channel. It remembers context across threads and days within a channel. Owners can view, edit and delete that memory.
• It decides when to speak. Since August 13 it reads the whole channel to decide whether to chip in, and goes quiet in channels where it has nothing to add. Any member can switch "Respond automatically" off.

Three ways in: tag it in a channel, DM it, or open it from Slack's AI assistant panel. Only the first is billed to the company.

Who gets it, and who pays for what

The billing split, from Anthropic's help center:

• @Claude in a channel runs under the company's Claude identity and is billed to the organization's usage balance. Anthropic's Enterprise guide says per-user and group limits don't cap it.
• A DM with @Claude runs on that person's own Claude account and draws on their seat, the same as using Claude anywhere else.
• The extra channel context Claude reads to decide when to speak doesn't count toward usage or spend limits on any plan, per Anthropic's August 13 post.

If your team is already on the Claude Team plan, there's nothing new to buy. There's a new line on the bill, and it grows with how often people tag Claude, not with how many people you have.

What a request costs, worked at the published rates

Anthropic doesn't publish a per-request Claude Tag price, and it doesn't say which model each Tag task uses. What it does publish is model pricing: Claude Sonnet 5.5 is $2 per million input tokens and $10 per million output tokens; Opus 5.5 is $4 and $20. Enterprise usage is billed at API rates. So here's what three everyday requests come to at those rates. The token counts are my estimates for a typical small-company Slack, not Anthropic's numbers.

Now a month. A 20-person company with three channels on Claude Tag, ten tagged requests a working day (mostly catch-ups and quick lookups at roughly $0.10 each), plus a weekly digest in each channel: 220 requests is about $22, the digests add about $2, so call it $25 a month on Sonnet. That's roughly one more Team seat, which is cheap for a teammate the whole company can ask. Double it if people lean on Opus or hand it long tool-heavy jobs.

Where it gets expensive is the same place it does in Claude Code or Cowork: agent work. A channel with standing instructions to triage every incoming bug report, read the logs and draft a fix runs many turns per task, and nobody is watching each one. That's the channel that needs its own limit.

The three settings I'd set before anyone types @Claude

1. An organization-wide Claude Tag spend cap. It's a hard cap: spend can't exceed it. Add per-channel limits for any channel with standing instructions. Admins get alerts at 75% and 95%, and work that would go over is declined, not cut off halfway. The same idea is behind the spend controls on Claude Enterprise pricing, but Tag has its own limit, separate from seat limits.
2. Member access. The default choices are anyone in the Slack workspace or any member of your Claude organization. If you have contractors or a shared Slack with clients, pick the narrower one. Role-based access is Enterprise only. Channels with Slack guests default to "Restrict", which blocks Claude there entirely. Leave it that way until you've decided otherwise.
3. Tools per channel, not company-wide. Access is set at three levels: organization, workspace, private channel. Put the sensitive connections (finance, HR, client files) in one private channel, and keep company-wide access to things everyone already sees. Since September, people can also bring their own connectors into a channel request on Team, with a review step before anything posts.

One more thing to know before setup day: only the Primary Owner or an Owner can set up Claude Tag. Someone with the Admin role can't. If the person running IT for you isn't an Owner in Claude, sort that out first.

What about Microsoft Teams?

If your company runs on Microsoft Teams instead of Slack, Claude Tag isn't there yet: Anthropic lists Teams as coming soon, with a waitlist. Until then, the Microsoft side is the Office add-ins in Claude for Microsoft 365. What makes Tag different from a chat window is reach: it works under its own identity in the tools you connect it to (GitHub, the CRM, Drive), so it can do the work in the thread, not only talk about it. Where Tag fits in the rest of the lineup is on Claude AI pricing.

Should a small business turn it on?

If your team already lives in Slack and you're on Claude Team, yes. Start with one channel where people constantly ask the same questions ("what's the status of X", "what did we decide about Y"). Set a small per-channel limit, run it for two weeks, and read the per-channel spend breakdown before you add more.

If you're on Pro or Max, Tag isn't an option, and moving five people to Team just for Slack is the wrong reason to switch plans. Move for the shared projects and the data settings, and treat Tag as a bonus.

Today's micro-action: scroll your busiest Slack channel and count how many messages in the last week were someone asking for a status, a number or a decision that's already written down somewhere. That count is what Claude Tag would answer. If you'd rather have the channels, permissions and spend caps set up properly for the whole team in one go, that's part of the AI implementation work I do.

Sources

Anthropic, Claude Tag product page (claude.com/product/tag); Anthropic Help Center, "What is Claude Tag?" (support.claude.com, article 15594475, dated September 24, 2026); Anthropic Help Center, "Claude Enterprise consumption guide"; Anthropic, "Claude Tag now reads even more of the room" (August 13, 2026) and "Claude Tag now supports personal connectors in channels" (September 24, 2026); Anthropic, claude.com/pricing (Team and Enterprise seat prices); Anthropic, Claude API pricing (platform.claude.com, Sonnet 5.5 and Opus 5.5 rates). All read October 25, 2026. Prices in USD. The per-request token counts and the 20-person month are my estimates, costed at those published rates.