25 October 2026 · Devarshi Mishra

Salesforce will bill Claude as its own line item. Map your AI meters before your next renewal

The Information reports that Salesforce will charge customers a separate fee to run Anthropic's Claude inside its apps, with no rate card published yet. Here's how I'd map every AI meter in an org, model by model, before that fee shows up on an order form.

• Salesforce
• Claude
• Agentforce
• Pricing
• Flex Credits
• RevOps

Salesforce customers have been buying AI the way they buy most things from Salesforce: a seat, maybe a premium edition, and some credits that someone in finance tracks once a quarter. That is starting to change. On 22 October The Information reported that Salesforce will charge customers a separate fee to run Anthropic's Claude inside its apps. The article is paywalled, so I'm relying on the coverage that followed it, mainly The Inference and SaasRise. Both say the same thing: no rate card has been published, and nobody knows yet how big the fee is or who absorbs it.

I'm not going to guess the price. What I can say is what this means for anyone who runs a Salesforce org. The model is becoming a line you can see, and once you can see it, someone will ask you to justify it. So my advice is simple: map every AI meter in your org, model by model, before that line shows up on an order form.

What has actually been reported

Here is what I could confirm from sources I read:

• The fee: a separate charge for running Claude inside Salesforce's apps, reported by The Information on 22 October. The Inference notes there is no published rate card.
• The market reaction: The Inference and SaasRise both report that Salesforce shares rose 3.5% on the day while the Nasdaq fell.
• The adoption gap: The Inference reports that only 5% of sales and service users on Salesforce's products have paid for the premium AI editions, which sell at a 60 to 80% premium, while Agentforce annual recurring revenue reached $1.5 billion.

SaasRise's headline says Salesforce "announced" the fee. I couldn't find a Salesforce announcement or price list, so I'm treating it as a report until a Salesforce document or an order form says otherwise.

You already pay per model, you just don't see it

This isn't the first time model choice has changed the bill. Salesforce's own Flex Credit rate card FAQ explains that prompt usage is metered by the model you pick. Prompts are grouped into usage types such as Basic, Standard and Advanced, each call includes 2,000 tokens, and larger calls trigger extra multipliers. The FAQ lists Claude 4 Sonnet in the Standard tier and Claude 3 Haiku in the Basic tier. In other words, picking a model in Prompt Builder or an agent has been a pricing decision for a year. Most teams I talk to never treated it like one.

Then there is Claude outside Salesforce. When Claudeforce launched in August, Patrick Stokes, Salesforce's president of applications and marketing, told VentureBeat that customers pay Salesforce through headless consumption pricing tied to their user license edition, and contract separately with Anthropic for the Claude inference. His words: "You can't buy this on one piece of paper at the moment." The D*AI*LY Brief added the integration detail I care about: MCP tool calls draw on the same daily API allocation as your existing integrations.

So depending on how your users reach Claude, you may already have three meters running: Flex Credits for prompts and actions, your org's API limits, and an Anthropic contract. The reported fee could add a fourth, or replace one of them. Either way, you can't budget for it if you don't know where Claude runs today.

Where I'd look first

Most of this sits in places admins rarely open together.

Start with Digital Wallet. Salesforce's FAQ says it shows your Flex Credit and Einstein Request balances, and usage over what you bought is billed monthly in arrears. That last part is the important one. There is no automatic stop. If an agent runs more than you planned, you find out on the invoice.

Next, list every prompt template, agent and flow that calls a model, and write down the model it uses. I'd add a column for the business owner, because when a model starts costing more, someone has to decide whether that workflow is worth it. That decision belongs to the person who owns the outcome, not to the admin who built it.

Then check the paths that bypass the UI. Claude connected over MCP, Slack features, and any integration your developers wired up during a pilot. These are the ones that grow quietly, because nobody sees them in a Lightning page.

What I'd do before your next renewal

1. List every AI feature that calls a model. Prompt templates, agents, flows and Slack features, with the model each one uses.
2. Tag each one with a business owner. The person who can say whether the outcome is worth the cost.
3. Pull 90 days of Digital Wallet usage. Find the top five consumers of Flex Credits.
4. Move simple jobs to smaller models. Test whether summaries and classification really need a Standard tier model.
5. Check API headroom for MCP and agents. They share the daily limit with your integrations.
6. Find any separate Anthropic contract. Know who signed it and what it caps.
7. Ask your account executive in writing. How will the Claude fee be priced, capped and reported?

The takeaway for CRM and RevOps teams

A visible model fee isn't bad news on its own. It's a chance to see what AI actually costs per workflow, which most teams can't answer today. The risk is meeting the fee for the first time at renewal, with no list of where Claude runs, no owner for each workflow and no usage history. An inventory, an owner per workflow and a written question to your account executive will put you in a much better position, whatever the rate card ends up saying.