The figures in this topic are illustrative estimates and don’t constitute a quote, offer, commitment, or guarantee of pricing or consumption. Actual Flex Credit usage depends on the number of prompts, models selected, token volume, and context sent with each request, and will vary. Rates, model availability, and product features are subject to change without notice. Confirm all figures and terms with your Salesforce account team before making any purchasing decision. Nothing in this topic creates a binding obligation on Salesforce.
Warning
Estimate Flex Credit consumption for Agentforce Vibes on metered (Flex Credit) billing. Use the single-prompt examples to understand what individual prompts cost, or use the team estimator to project monthly spend.
Three rules define every charge. Together, they mean your bill is a function of two variables: how many tokens your prompts consume, and which model processes them.
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. All tokens count: input, output, and cache read and writes, at the same rate.
| Tokens | Units |
|---|---|
| 6,500 tokens | 4 units |
| 2,000 tokens | 1 unit |
| 500 tokens | 1 unit |
See Metered Usage Via Flex Credits for more information.
Every model in Vibes maps to a fixed Prompt Usage Type: Starter, Basic, Standard, or Advanced. Sonnet and GPT are Standard. Opus is Advanced. Switching the model changes the type, and the type is what drives the multiplier in Rule 3.
| Model | Prompt Usage Type |
|---|---|
| Sonnet 4.6 | Standard |
| GPT-5.4 | Standard |
| Opus 4.8 | Advanced |
See Large Language Model Support for more information.
Each Prompt Usage Type has a published Flex Credit multiplier on the Rate Card. That multiplier, times the number of units, is what Salesforce charges. Advanced is four times the credit cost of Standard for the same token volume.
| Prompt Usage Type | Multiplier |
|---|---|
| Starter | ×2 |
| Basic | ×2 |
| Standard | ×4 |
| Advanced | ×16 |
See Flex Credits Rate Card (June 17, 2026) for more information.
1credits = ceil(tokens / 2000) × multiplierExample: 6,500-token Sonnet prompt = ceil(6500 / 2000) × 4 = 4 × 4 = 16 credits
The table shows real token totals from a single Agentforce Vibes session: one prompt, no deployment step. The prompt asked Vibes to create a Project__c custom object with auto number, name, start date, and number-of-days fields; a bulk-safe ProjectTrigger Apex trigger to set the start date on insert; and a full Apex test class. The run produced approximately 7–12 files locally.
Token totals are from the Salesforce Flex Billing view B (cache-inclusive: all input, output, cache read, and cache write tokens counted). Deployment adds more tokens not reflected here.
| Model | Prompt Usage Type | Billable Tokens | Units | Flex Credits |
|---|---|---|---|---|
| Opus 4.8 | Advanced (×16) | 2,881,624 | 1,441 | 23,056 |
| Sonnet 4.6 | Standard (×4) | 1,371,177 | 686 | 2,744 |
| GPT-5.4 | Standard (×4) | 700,834 | 351 | 1,404 |
Sonnet 4.6 and GPT-5.4 are both Standard-tier models but produced different token totals for the same prompt. Sonnet primes and reads its own cache on each step. GPT-5.4 benefits from a warm gateway cache. Token volume isn’t uniform across Standard-tier models.
Use the inputs with the formula to estimate monthly Flex Credit consumption.
| Input | Default Value |
|---|---|
| Developers using Vibes | 10 |
| Working days per month | 21 |
| Prompts per developer per day | 60 |
| Average tokens per prompt | 6,500 |
| Primary model | Sonnet 4.6 (Standard) |
| Preset | Avg Tokens | Typical Prompts |
|---|---|---|
| Light | 2,000 | Explain this Apex method; Write a Salesforce Object Query Language (SOQL) query; Add a JSDoc comment. |
| Typical | 6,500 | Generate a Lightning Web Component (LWC) with a wired Apex controller; Write a unit test; Refactor this method. |
| Heavy | 15,000 | Build a flow and Apex action across 3 files; Review a PR diff; Generate a test class with multiple bulk scenarios. |
| Agentic | 40,000 | Implement this user story end-to-end”. “Find and fix the failing tests in this module”; “Migrate this component to the new API and update callers. |
Use these formulas to calculate monthly Flex Credit consumption from your inputs.
1billable_tokens = input + output + cache_read + cache_write (all tokens billed)
2units/prompt = ceil(billable_tokens / 2000)
3prompts/month = developers × prompts/day × working_days
4credits/month = units/prompt × prompts/month × rate_multiplierThe following table shows the Flex Credit calculation for the default scenario.
| Metric | Value |
|---|---|
| Units per prompt | 4 units (ceil(6500 / 2000)) |
| Prompts per month | 12,600 (10 × 60 × 21) |
| Billing units per month | 50,400 |
| Prompt usage type | Standard |
| Rate multiplier | ×4 |
| Flex Credits per month | 201,600 |
All tokens processed count toward the 2,000-token unit: input, output, and cache-read and cache-write. There’s no caching discount, so Salesforce bills cached context at the same rate as fresh input. The measured token totals above already reflect this.
Note
Sandbox usage draws from the associated production org’s Flex entitlement. There’s no separate sandbox pool, and Salesforce bills sandbox prompts at the same multiplier as production. Digital Wallet flags whether consumption originated in sandbox vs. production.
Note
The factors below are ordered by how much they typically move the number, with the first three being the biggest surprises on a real bill.
Conversation history and prior turns. In a multi-turn session, Agentforce Vibes re-sends every earlier user message and assistant response on each new turn. Usually the largest driver in agentic runs: a 10-minute conversation costs far more than 10 times the first turn.
Tool call results. When Vibes reads a file, greps the repo, runs a test, or executes a command, the output lands back in context for the next turn. A test log, stack trace, or grep across a large module can add thousands of tokens per tool call.
Reasoning and thinking tokens. Opus generates internal reasoning tokens that count as output but don’t appear in the reply. A hard problem can produce many multiples of the visible answer in thinking tokens.
Retries and self-correction. When a tool call errors or the model rewrites its own output, each attempt is a billable turn. Agentic runs iterate. Plan for retries in the average.
System prompt and Vibes instructions. Fixed overhead sent on every request: the Vibes agent’s operating instructions, tool definitions, and formatting rules. Base cost before the user types anything.
Project-level instructions. Agentforce Vibes loads any custom rules file (project guidelines, user preferences, coding standards) into every request.
Retrieved context and grounding results. When Vibes pulls from the Salesforce Catalog or runs semantic retrieval against org metadata, those results become input tokens.
Open files, selected code, and multi-modal input. Files open in the editor, selected code passed as context, and any images or screenshots attached to the prompt all count as input tokens.
Cache read and write tokens. Agentforce Vibes caches the stable portion of context (system prompt, prior turns, large files) to speed later turns. Salesforce bills all cache read and write tokens at the standard rate. Caching improves speed, not cost.
Actual mix varies by session, task, and model.
Measured token totals are from single-turn Agentforce Vibes sessions captured in Billing view B (cache-inclusive; Salesforce Flex convention). No measured run included a deployment step.
Confirm rates with your Salesforce account team before making purchasing decisions. The Agentforce Vibes stock-keeping unit (SKU) is ineligible for courtesy concessions.
See Metered Usage Via Flex Credits for billing unit rules and token counting behavior.
See Flex Credits Usage for prompt usage type definitions.
See Large Language Model Support for the model-to-prompt-usage-type mapping.
See Agentforce and Data 360 Rates for the current Flex Credits Rate Card and multipliers.
j