Tokens
What is the difference between Salesforce Flex Credits and tokens?
Independent analysis, not affiliated with or endorsed by Salesforce. Built on 11 Salesforce pages, each cited below and checked on 2 Nov 2026.
Quick answer
LiveLast verified 2 Nov 2026
Flex Credits are the currency Salesforce bills AI in, at $0.005 each at list; tokens are the units of text an AI model reads and writes. An agent action costs a flat 20 credits whatever its tokens. Only prompts bill by tokens: each started 2,000-token unit costs 2, 4 or 16 credits, by model tier.1–3
One call · 6,500 tokens in and out2,000 tokens per prompt unit, rounded up
2,000
Unit 1
2,000
Unit 2
2,000
Unit 3
500 of 2,000
Unit 4 · billed in full
The same call in credits, and one agent actionCalculatedrate card × units
- Starter prompt2 a unit8 credits · $0.04
- Basic prompt2 a unit8 credits · $0.04
- Standard prompt4 a unit16 credits · $0.08
- Advanced prompt16 a unit64 credits · $0.32
- Agent actionflat, any tokens20 credits · $0.10
Units × the tier's credits a unit; dollars at the list price of $0.005 per credit. An action is one unit however many tokens it processes.
One unit you pay in, one unit the model counts
Flex Credits are Salesforce's billing currency for Agentforce and Data 360, at $500 per 100,000 credits at list. Tokens are not a price. Salesforce defines them as the units of data an AI model processes, input and output together, and uses them only to size some meters; the bill itself is in Flex Credits.2,4
An agent action is where most people expect tokens to matter, and they don't. Salesforce Help: "Currently, every standard action called by the agent is metered as one standard action, irrespective of the number of tokens processed." That is 20 credits a run in production, $0.10 at list, for a short prompt or a long one.2,3
Prompts are where tokens count. Embedded AI that calls a model directly, such as Service Replies, is metered in prompt units of 2,000 tokens, input plus output, rounded up per call: a 6,500-token call is metered as 4 prompts. A unit costs 2 credits at the Starter and Basic tiers, 4 at Standard and 16 at Advanced, following the model actually used at run time.2,3
When an agent runs a prompt template, it is billed as an action, not as prompts. Orgs on older contracts count the same embedded AI in Einstein Requests instead, at 4, 4, 10 or 38 per 2,000-token block, and Salesforce publishes no dollar price per Einstein Request.2,5
Agentforce Vibes, Salesforce's tool for developers, bills each model request in the same prompt units, with input, output and cache tokens all counted at the same rate. Salesforce's own example: one prompt on an Advanced-tier model used 2,881,624 tokens, or 1,441 units and 23,056 credits.6
Everything else ignores tokens: voice bills per action or per hour, Data 360 AI processing per megabyte, and the announced meter for external agents' MCP and API calls, the Headless Platform Interaction, will bill per call. Its multiplier is TBA and it isn't metered yet.2,3,7
Which meters count tokens
Salesforce's rates, whether tokens change the charge, and one unit priced at list.
| Meter | Rate | Do tokens matter? | At list | Status |
|---|---|---|---|---|
| Agent action, text3 | 20 credits per action | No · flat per action | $0.10per action | Live |
| Agent action, voice3 | 30 credits per voice action | No · flat per action | $0.15per action | Live |
| Starter prompt3 | 2 credits per Starter Prompt (per 2,000 tokens) | Yes · per started 2,000 tokens | $0.01per unit | Live |
| Basic prompt3 | 2 credits per Basic Prompt (per 2,000 tokens) | Yes · per started 2,000 tokens | $0.01per unit | Live |
| Standard prompt3 | 4 credits per Standard Prompt (per 2,000 tokens) | Yes · per started 2,000 tokens | $0.02per unit | Live |
| Advanced prompt3 | 16 credits per Advanced Prompt (per 2,000 tokens) | Yes · per started 2,000 tokens | $0.08per unit | Live |
| Agentforce Vibes6 | Prompt units of 2,000 tokens · cache tokens count in full | Yes · cache tokens count in full | By unit, see rate | Live |
| Einstein Requests, older contracts5 | Starter 4 · Basic 4 · Standard 10 · Advanced 38 per 2,000 tokens | Yes · per 2,000-token block | By unit, see rate | Live |
| Speech and translation3 | Speech-to-text 150 credits per hour · text-to-speech 6,000 per 1M characters · translation 4,000 per 1M characters | No · per hour or per characters | By unit, see rate | Live |
| Agents' MCP and API calls (announced)3 | TBA | No · per call | TBA |
Source. Salesforce's rate cards, pricing pages and Help articles; one unit at list is $0.005 per credit × the rate, a calculation. Every prompt tier counts input plus output tokens in 2,000-token blocks, rounded up: a 6,500-token call is metered as 4 prompts. An action is billed each time it runs, whether it was triggered deterministically or chosen by the LLM, and counts as one unit however many tokens it processes.
Last verified 2 Nov 2026
What makes a token bill grow
Four things move a token-based charge. Agent actions are not among them.
1. Longer prompts and longer answersLive
Input and output both count, so a prompt bills one more unit every time it passes another 2,000 tokens.2
2. A more expensive model tierLive
The same unit costs 2 credits on a Basic-tier model and 16 on an Advanced one; the charge follows the model actually used at run time.2,3
3. Rounding up, per callLive
Every call bills whole units: a 500-token call still bills one full unit, and a 6,500-token call is metered as 4 prompts.2
4. Long coding sessions in Agentforce VibesLive
In Salesforce's example, one prompt with no deployment used 700,834 to 2,881,624 tokens depending on the model: 1,404 to 23,056 credits. The unmetered Vibes offer caps premium models at 2,500 requests or 36M tokens per user per org every 30 days.6,9
Prompts priced from the rate card
Salesforce's own rounding example and its Agentforce Vibes session, worked through at list. Arithmetic on published rates.
| Model tier | Credits a unit | This call, 4 units | At list | Einstein Requests, older contracts |
|---|---|---|---|---|
| Starter | 2 | 8 | $0.04 | 16 |
| Basic | 2 | 8 | $0.04 | 16 |
| Standard | 4 | 16 | $0.08 | 40 |
| Advanced | 16 | 64 | $0.32 | 152 |
Source. Salesforce Help: Every prompt tier counts input plus output tokens in 2,000-token blocks, rounded up: a 6,500-token call is metered as 4 prompts. Einstein Requests: Starter 4 · Basic 4 · Standard 10 · Advanced 38 per 2,000 tokens; Salesforce publishes no dollar price for them. Arithmetic by CalculateForce.
Last verified 2 Nov 2026
The call needs 3.25 units of 2,000 tokens, so it bills 4: the last unit holds 500 tokens and is billed in full. The same call run by an agent as a prompt-template action would be one action at 20 credits per action.
Salesforce's Agentforce Vibes example: one prompt, three models
| Model tier | Credits | Note |
|---|---|---|
| Advanced tier | 23,056 · $115 | |
| Standard tier | 2,744 · $14 | |
| Standard tier | 1,404 · $7.02 |
Source. Salesforce's Agentforce Vibes licensing guide; Salesforce calls the figures illustrative estimates. Its page names the models; this registry names none. Dollars at $0.005 per credit, a calculation. Last verified 2 Nov 2026
Turn tokens into Flex Credits
Tokens a call, calls a month and the model tier in; units, credits and dollars at list out. Arithmetic on the rate card.
6,500 tokens ÷ 2,000, rounded up
4 units × 4 credits (Standard) × 10,000 calls
160,000 credits × $0.005 per credit
The same month by tier, and as agent actionsCalculatedrate card × your volumes
- Starter2 a unit80,000 · $400
- Basic2 a unit80,000 · $400
- Standard4 a unit160,000 · $800
- Advanced16 a unit640,000 · $3,200
- As agent actions20 a run, any tokens200,000 · $1,000
Rounding up bills 1,500 tokens a call that the last unit could still have held. 2,000 tokens per prompt unit, rounded up.
Prompt meters apply to embedded AI that calls the LLM directly, such as Service Replies; when an agent runs a prompt-template action, it is billed as an action, not as prompts.
What Salesforce publishes about tokens, and what it doesn't
Published
- An action is one unit however many tokens it processes.2
- Prompts: 2,000 tokens per unit, input plus output, rounded up per call.2
- Prompt rates: 2, 2, 4 and 16 credits per unit, Starter to Advanced.3
- Token counts are recorded in the AI usage data model, metered or not.10
- A rough guide: 2,000 tokens are about 1,500 words.5
- Older contracts: Einstein Requests at 4, 4, 10 or 38 per 2,000-token block.5
Not published
A dollar price per Einstein Request, or a fixed conversion to Flex Credits5
Not published · an Order Form may state a contract priceCredits per Headless Platform Interaction, the meter for agents' MCP and API calls3
Flex Credits Rate Card · TBAHow many tokens your own prompts use10
AI usage data model · token usage per callYour contracted price per credit1
Order Form
Whose bill depends on tokens
Tokens decide a charge only where Salesforce meters prompts: embedded AI that calls a model directly, Agentforce Vibes, and Einstein Requests on older contracts.2,5,6
- Service teams using embedded AI such as Service Replies, which calls a model directly and bills prompt units.2
- Developers on metered Agentforce Vibes, where every model request bills prompt units, cache tokens included.6
- Orgs on older contracts, whose embedded AI and Models API calls are metered in Einstein Requests rather than Flex Credits.5
- Anyone reconciling a Flex Credits line: Digital Wallet shows credits and units, while token counts sit in the AI usage data model.10
Where tokens don't matter
- Agent actions: 20 credits a run in text, however many tokens the action processes.2,3
- Voice, Data 360 AI processing and the announced per-call meter for external agents.2
- Other things called tokens: Revenue Cloud's token-based selling, a feature for selling your own products as prepaid tokens; OAuth, security and design tokens; and, outside Salesforce, Autodesk's Flex tokens.11
08Next action
Most agent bills are actions, not tokens
The Agentforce calculator prices a month of agent actions at 20 credits each, with prompts and voice beside them.
What the instrument adds
Rate card
Each meter at Salesforce's published rate.
Your month
Credits and dollars for your own volumes.
Your rate
Your contracted price per credit, from the Order Form.
Where every line comes from
Salesforce's own pages. Each superscript in the article points to one of them; the excerpts are verbatim.
- 01
Salesforce Introduces New Flexible Agentforce Pricing to Accelerate the Digital Labor Revolution
Salesforce · 15 May 2025
“Each of these actions consumes just 20 Flex Credits ($0.10 per action)”
- 02
Flex Credits Billable Usage Types
Salesforce Help · Accessed 2 Nov 2026
“Agent actions are billed when they're run, whether they're run deterministically (for example, in reasoning instructions) or during agent reasoning with an LLM.”
- 03
Flex Credits Rate Card (updated as of August 31, 2026)
Salesforce · 31 Aug 2026
“Usage types, tiers, and associated multipliers may be updated from time to time. Changes to tiers will be treated as multiplier changes.”
- 04
Salesforce · Accessed 2 Nov 2026
“Flex Credits are a new unit of payment for Agentforce that align cost to the business value your AI agents create.”
- 05
Einstein Requests Rate Card (updated as of September 21, 2026)
Salesforce · 21 Sep 2026
“Generally, a 100-page text-heavy PDF document might contain around 50,000 to 75,000 tokens, while a single high-resolution image could be represented by several thousand tokens in a vision-language model.”
- 06
Estimate Flex Credits Usage (Agentforce Vibes Developer Guide)
Salesforce Developers · Accessed 2 Nov 2026
“Agentforce Vibes bills every request to a model in Prompt Usage Types. One Prompt Usage Type covers up to 2,000 tokens processed, rounded up.”
- 07
Understand AIforce Impact (Knowledge Article 005360285)
Salesforce Help · 17 Sep 2026
“Every successful call a registered agent makes to Salesforce—whether it arrives over Model Context Protocol (MCP) or direct API—is metered as a Headless Platform Interaction (HPI)”
- 08
Introducing Flex Credits for Agentforce (Salesforce Partner Community)
Salesforce · Undated, launch-era (2025)
“There will be a set amount we enforce of 10,000 tokens per action. Excess would consume additional Flex Credits.”
- 09
Agentforce Vibes Billing (Agentforce Vibes Developer Guide)
Salesforce Developers · Accessed 2 Nov 2026
“Limit for premium models is 2,500 requests or 36 million tokens, per user per org, whichever is reached first. Limits reset every 30 days after first use.”
- 10
Metering for Agentforce and Generative AI Usage
Salesforce Help · Accessed 25 Oct 2026
“The usage data model captures all generative AI and Agentforce usage including unmetered usage, reason for unmetering and token usage.”
- 11
Configure Token Resources for Flexible Selling Models (Trailhead)
Trailhead · Accessed 2 Nov 2026
“Token-based selling is a flexible consumption model where customers purchase virtual tokens in advance and redeem them across a range of products or services.”
Last verified
2 Nov 2026
Every Salesforce price, rate and status on this page comes from CalculateForce's pricing registry, checked against the pages above on 2 Nov 2026. Salesforce changes its rate card several times a year; where a value is announced or TBA, the page says so.
Independent. Not affiliated with or endorsed by Salesforce. Spotted a change? Tell us.
Frequently asked questions
What is the difference between Flex Credits and tokens?
Flex Credits are what Salesforce bills you in, $0.005 each at list. Tokens are the units of text an AI model processes, input and output together. Salesforce uses tokens only to size some meters, mainly prompts; the bill itself is always in Flex Credits.1,2
How many tokens is one Flex Credit?
There is no single rate, because only prompts are token-based. With every 2,000-token unit full, one credit covers up to 1,000 tokens at the Starter and Basic tiers, 500 at Standard and 125 at Advanced (CALCULATED). Rounding up per call means real prompts get less. An agent action ignores tokens: 20 credits a run.2,3
Does an Agentforce action cost more if it uses more tokens?
No. Salesforce Help says every standard action an agent calls is metered as one standard action, "irrespective of the number of tokens processed": 20 credits in production, $0.10 at list. Salesforce's wording is "currently", so the rule can change.2,3
Einstein Requests vs Flex Credits: what is the difference?
Einstein Requests are the older unit. Orgs on older contracts meter embedded AI and Models API calls in them, at 4, 4, 10 or 38 per 2,000-token block by tier; orgs with Einstein Request to Flex fungibility are charged in Flex Credits. Salesforce publishes no dollar price per Einstein Request, so a legacy bill can't be priced from public sources.5
Where do I see token usage in Salesforce?
In the Agentforce and generative AI usage data model, which records every AI event, metered or not, with its token usage. Digital Wallet shows credits and units, not tokens: its Units Consumed is back-calculated from rounded credits.10
Are MCP and API calls charged by tokens?
No. The meter Salesforce has announced for external agents' MCP and API calls, the Headless Platform Interaction, will bill per call in Flex Credits. Its multiplier is TBA and it isn't metered yet; Salesforce promises 30 days' notice before metering starts.3,7
How many words is 2,000 tokens?
About 1,500 words on average, by Salesforce's own rough guide; a 100-page text-heavy PDF is around 50,000 to 75,000 tokens. The real count depends on the model, the language and the formatting.5
What does a million tokens cost in Salesforce?
As prompts, in full 2,000-token units at list: $5 at the Starter and Basic tiers, $10 at Standard and $40 at Advanced (CALCULATED from 2, 4 and 16 credits a unit at $0.005). Rounding per call adds to that; agent actions don't bill by tokens at all.1–3
Is Revenue Cloud's token-based selling the same thing?
No. Token-based selling is a Revenue Cloud feature for selling your own products as prepaid tokens your customers redeem. It is not how Salesforce bills you and has nothing to do with an AI model's tokens; neither do Autodesk's Flex tokens.11
Keep reading
Pricing Alerts
Notify me when Salesforce changes this.