Salesforce Agentforce Pricing in 2026: What You’ll Actually Pay

Salesforce Agentforce has been in the spotlight for quite some time because of its autonomous actions across the Salesforce ecosystem. Its pricing has become one of the most talked-about topics and most misunderstood throughout the year. 

Unlike other per-seat license models, Agentforce pricing is structured around consumption. This means the amount printed in your subscription package infrequently matches the number on your invoice. 

Salesforce announced four different Agentforce pricing models in two years, including conversational, flex credits, per-user licensing, and a recently announced pay-per-resolution model.  All 4 pricing models are currently live for all users. The recently launched pay-per-resolution model is an outcome-based pricing structure where users are only liable to pay when an AI help agent autonomously resolves a customer issue from start to finish at a $2/ conversation fee. 

If you’re looking to budget for Agentforce in 2026, make sure to understand which pricing model applies to your org’s use case and what it doesn’t cover. 

This blog breaks down agentforce pricing published on Salesforce’s latest page. We’ll also uncover the hidden costs and potential price implications for your org. 

Let’s discuss why Agentforce pricing feels complicated to most enterprises. 

Why Salesforce Agentforce pricing looks complicated

In most cases of enterprise software, organizations have one pricing lever: seats (number of users allowed access). agentforce is different because it isn’t sold as a seat-based product; rather, it is sold on a consumption basis. 

Salesforce’s own pricing page describes it as flexible pricing that you can scale on your terms. It is offered through consumption-based flex credits or conversations, or through per-user licensing. 

Because of the offered flexibility, the pricing feels so confusing, and enterprises are unable to assess their consumption. For example, a company with 60 or more employees running a high-volume customer service bot could easily outspend a 500-person company that uses Agentforce for internal Q&A use. 

Agentforce tracks the bill based on usage, not headcounts. So the listed price on the official website tells part of the story until you track it against your actual workflow usage. To better understand Agentforce’s pricing model, have a look at how it’s listed on the official Salesforce website. 

Salesforce Agentforce pricing in 2026: Current model pricing at a glance 

As per Salesforce’s official pricing page, there are seven ways to access Ageneforce, spanning a free tier, two consumption-based models, three licensing models, and one per-resolution-based model. Check the Agentforce rate card to get a full understanding of how each model is billed and how you can optimise your consumption within limits. 

How Agentforce flex credits work

Salesforce launched flex credits as a consumption-based currency, and they are the most flexible for users. It is the most misunderstood part of Agentforce pricing. According to Salesforce, every standard agentforce action costs 20 flex credits, and each Agentforce voice action costs 30 flex credits. 

It is priced at $500 per 100000 credits, which amounts to roughly $0.10 per standard action and $0.15 per voice action. Check your Agentforce flex credits roadmap before signing up.

Here’s the catch: an action mentioned here is granular. Authenticating a customer record, pulling his case history, and drafting a detailed response each can be separate actions. In this case, a single customer conversation might consume anywhere from one to dozens of actions, depending on the complexity of the case. 

Let’s take a look at some realistic examples that are worth quoting, as they reflect how fast a small per-action cost during your conversation can compound at scale. 

• HR benefits helpdesk agent case– 50 employees asking questions/day. 2 actions for each employee amount to 100 actions per day – $300/month. 
• E-commerce refund and return-  Company gets 500 return requests per month, which amounts to 4 actions each (2000 actions/month)- $200/month.
• Generating a sales quote- 20 reps in the Salesforce ecosystem generating 5 quotes/day. This would make to 6 actions per quote and amounting to 600 actions/month- $ 1200/month.   
• Knowing order status with self-service bot- Org receives 20 where is my order requests/ day. This leads to 2 actions each for every order, using 40 credits – $120/month.
• Voice reservation changes- Receiving 300 calls/month that lead to 4 actions/call – $180/month. 

All calculations are made on the official rate of 20 flex credits at $0.10 per standard action, as mentioned on the Salesforce Agentforce pricing page (2026). The actual action count might vary depending on the agent design and the complexity of each conversation. 

Agentforce help agent: Pay-Per-Resolution 

Salesforce had launched three pricing models, and in june 2026, salesforce added a fourth option designed around per-conversation resolution. On June 25, 2026, Salesforce announced Agentforce Agent help, which is a pre-packaged autonomous customer service agent priced at $2 per resolution. 

The mode reached general availability in July, 2026.  It is an outcome-based pricing approach where users are only charged when the AI agent resolves an issue from start to finish. This pricing model is best suited for organizations with high volume and repetitive customer support requests that want to attach their outsourcing cost to successful end-to-end outcomes.

What counts as an action 

An action in the flex credit model relates to the step the agent takes. Its common examples include:

• Searching into a knowledge base or a document index
• Sending an email or message 
• Calling an apex call, external API or flow 
• Escalating to a human agent 
• Querying a Salesforce object, like reading a case record 

A single end-to-end user interaction might involve multiple actions tied together. 

“Where is my order?” might trigger the following actions 

• Authenticate user 
• Query order object 
• generate a natural language response 

Totalling 3 actions that amount to 60 credits = $0.30

Hidden variables: Token overflow inside single action

Here are some details that you rarely find on a vendor’s pricing page. But from a practical perspective, it matters a lot in production. Salesforce caps each standard action at 10,000 tokens of processing. This cap is set to make sure we stay under this limit, as it can lead to token overflow. When you go above the prescribed limit, Salesforce counts the same action multiple times. 

The threshold is evaluated per action and not pooled across conversations. So this is easy to miss until the bill is higher than expected.   

An agent action that pulls a concise and summarized snippet costs $0.10. The same conceptual 1 action that instead loads an entire policy PDF into context can quietly cost 3 times more, not because Salesforce changed the rate, but because the agent design has pushed it past the token limit. 

Keep the prompts and knowledge retrieval focused and summarised rather than dumping the entire document into the context. Make sure to use a digital wallet to track which actions are constantly overflowing. 

Flex credit vs. conversation: Find the real break-even point 

A standard action costs $0.10, and a single conversation costs $2.0. The math works out to a clear break-even point of 20 actions per interaction. Below that, flex credits win in any case, and above it, conversation wins. 

Mapped against a real use case, the gap is dramatic at the low end. 

This is exactly why an organization must not assume flex credit is always cheaper. A simple FAQ bot or case-management agent will always cost less under flex credits. But an SDR agent who sends an initial outreach email, followed by dozens of follow-up emails to the same lead, can blow past 20 actions per lead. At this point, a flat $2 conversation rate can be considered a better deal.

Salesforce doesn’t allow both models in a single org, and this is worth modelling with your actual agent design before you pick a model, not after. 

Conversation pricing: flat and limited

Salesforce launched conversation pricing as the original Agentforce billing model as a flat $2 per conversation, regardless of how many actions were triggered internally. It only applies to customer-facing agents, and Salesforce explicitly mentioned that conversation and flex pricing can not run in the same org at the same time. You can choose one pricing model at a time, which makes it an early and hard-to-reverse decision. 

Real-world cases: Pricing agentforce voice for contact center

Agentforce voice actions are an area where pricing gets layered. There isn’t one cost; there are three, and picking the wrong combination can double your cost without anyone noticing. Here are the layers that need your consideration.

Layer 1(AI credits)

Salesforce measures voice through two mutually exclusive models. Voice actions (30 flex credits per discrete action- $0.15), and voice minutes (60 flex credits per minute of active all time- $.30/minute in production). If both models are enabled, then voice minutes automatically take precedence. 

For example, a 3-minute call handling 4 actions (identify the caller, look up reservation, respond to question, and finalize update) costs $0.60 under the voice action, but it is billed at $0.90 under voice minutes. Means the right model depends on whether your calls are short and simple or longer with unpredictable duration. 

Voice action is preferred when a call requires fewer than two actions per minute of talk time. On the other hand, voice minutes win when one action density climbs higher than that.  

Layer 2 (Telephony platform)

On top of the AI credits, you’ll still need a phone connectivity licence through Service cloud voice, and it is separate from Agentforce entirely. Salesforce’s published rates run at $0.50/user/month for bring your own telephony. At the same time, $75 to $200/user/month for tiers of pre-integrated Amazon connect minutes.

Layer 3 (Data cloud/observebility credits), if agent analytics and observability features are enabled

In case the contact center puts together a 10-agent contact center that runs through 1000 AI-handled calls in a month. Under the voice action model, it would land $600/month in flex credits. Plus there would be a $1250 for a mid-tier Amazon connect telephony license before any data cloud observability costs. It’s a reasonable check for any team pricing out a voice deployment where the AI credits are the smallest of the three line items, not the biggest. 

Change in Salesforce Agentforce pricing: a quick timeline 

The history of the Agentforce explains why so many users of the ecosystem feel like they are chasing a moving target. Agentforce was launched in 2024 with a $ 2-per-conversation plan as the only option. In 2025, salesforce launched flex credits as a more granular and consumption-based alternative. 

Recently, Salesforce added the per-user licensing that includes a $5/month Agentforce user license and another $550+/momth Agentforce 1 edition. It provided the buyers with a flat-fee path for the company-wide employee-facing rollout. 

The launch of three live models in two years feels like unusually fast iteration for enterprise software, and it reflects something crucial. Even Salesforce is figuring out how to translate AI agent work into a predictable bill customers can rely on. You can observe that the iteration hasn’t stopped as Salesforce launched the pay-per-conversation model in mid 2026.

Here are two important rules worth remembering when negotiating on Agentforce pricing. 

• The unused flex credits do not roll over into the next month’s contract term, and in such a case, accurate forecasting is important. 
• There will be no punitive overuse penalty. If you exceed your entitlement, you’ll be billed monthly in arrears at the contracted rate. This can be tracked through Salesforce’s digital wallet. 

Key practical ways to keep Agentforce usage under control 

Beyond picking the right model for your business, here are some operational habits that can make a real difference to your monthly bill. 

• Track the consumption on the digital wallet, not in invoice- Salesforce, for its flex credit and other models, has launched a free digital wallet that shows real-time and action-level credits of the agents. Make sure to set alerts at 50%, 75% and 90% of the allotment catches that would ensure a misconfigured or looping agent before it consumes your monthly budget in a single day. 
• Test in Sandbox before real production- Sandbox actions are billed at a discounted rate versus production rates. It can be done to ensure the batch testing is done there before it ever touches a live customer in production. It is cheaper to test in Sandbox and catch expensive design flaws like token-heavy prompts at an early stage. 
• Design for token ceiling- Each action is capped at 10,000 tokens before it starts multiplying. Focus on concise prompts and summarised knowledge retrieval, and they are a direct lever on your Flex credit limit. 

Is agentforce worth it: a quick ROI check 

None of the investment matters if it doesn’t pay you back. Here are a few things to notice before signing up.

• Customer service deflection math justifies Agentforce fastest- If your average cost/ human handled support case runs $15 to $25 and you’re handling meaningful monthly volume, then AI deflection can produce realistic savings. But this calculation works only at scale. 
• Create special focus on Governance- As per Gartner’s research, over 40% of Agentic AI projects in the current scenario would be cancelled by the end of 2027. It is driven primarily by escalating high costs, unclear business vision, and inadequate control over costs. There’s no proof of technology failure, but because of ineffective governance, the businesses are unable to proceed with agentic AI technology.

If you’re planning to deploy Agentforce in the current Salesforce ecosystem, then get a quick Agentforce ROI check. Optimize your current Salesforce Agentforce business environment based on a realistic ROI check. 

Three questions to ask before you sign up 

Get a reality check by asking the right questions about your data readiness. 

1. Are your knowledge base and Salesforce data actually ready?

Poor input produces poor output, no matter how much you invest in the Salesforce ecosystem. 

2. Which pricing model actually matches your real consumption pattern?

Audit your existing usage of the agentic capabilities across different models. 

• High volume and simple customer interactions would match the conversation model.
• Mixed use cases across different departments would match flex credit.
• Company-wide internal roll-out often matches the licensing model.
• You can evaluate the Help agent by getting a clear and contractual definition of what would count as a billable resolution.

3. If usage patterns change, what’s the exit plan?

Flex credits and conversation model can not run simultaneously in the same org. Make sure to negotiate flexibility into your contract before signing, not after. 

The bottom line 

The Salesforce Agentforce pricing isn’t confusing by accident. it reflects a new pricing category as part of a consumption-based approach backed by mandatory data infrastructure requirements that most publishers leave out. Each model has its own variables: token overflow inside flex credits, layered voice billing, action destiny break-even against conversations, and an undefined notion of resolution in the new pay-per-resolution model. Every org, based on its consumption architecture and behavioural model, needs to choose between the models. 

The organization that gets realistic ROI from Agentforce is the one that models their actual volume before signing up. The price in data cloud and implementation aspects from day one matches the pricing model to the shape of their workflows rather than defaulting to whichever one on a sales deck leads. Make sure to validate results on a narrow pilot before committing to a company-wide adoption.  

Agentforce’s complex pricing approach is not understood by organisations that rely on website-published data. This is exactly why Cyntexa provides hands-on experience in scoping, implementing and optimising Agentforce deployment across service, sales, and voice use cases. It includes modelling flex credits vs. conversations vs. pay-per-resolution against clients’ real usage patterns. 

If you’re evaluating Salesforce Agentforce for your business, talk to our experts at Cyntexa about proposed pricing and a readiness assessment before your next renewal or pilot conversation.  

AUTHOR

Vishwajeet Srivastava

Salesforce Data Cloud, AI Products, ServiceNow, Product Engineering

Co-founder and CTO at Cyntexa also known as “VJ”. With 10+ years of experience and 22+ Salesforce certifications, he’s a seasoned expert in Salesforce Data Cloud & AI Products, Product Engineering, AWS, Google Cloud Platform, ServiceNow, and Managed Services. Known for blending strategic thinking with hands-on expertise, VJ is passionate about building scalable solutions that drive innovation, operational efficiency, and enterprise-wide transformation.