Summary

  • The rate card prints a production and a sandbox multiplier side by side; across the thirteen rows my read pairs, the sandbox column applies four-fifths on four, an identical multiplier on seven and zero on two, and the only general sandbox sentence in the document defines what a pre-production environment is, not which row earns which ratio.
  • Sandbox treatment does not follow the size of the meter: the largest multiplier in the table, 6,000 per million characters, is unchanged in pre-production, a 400-credit resolution falls to zero, and in sandbox a standard action and an advanced prompt both cost 16 credits although production separates them 20 to 16.

What the column prints

The governing sentence is one line long, and it makes the row label the thing the rate depends on: "Customer consumes Flex Credits at the applicable multiplier rate based on the Usage Type identified below." The document carrying it is titled "Working Version Flex Credits Rate Card 11.2026" and its footer reads "Updated as of November 3, 2026. Technical restrictions may apply." (Flex Credits Rate Card).

Below that sentence sit two numeric columns. The thirteen rows my read pairs against production and sandbox are these:

What the column prints
Usage type Production Sandbox
Standard Action 20 16
Custom Action 20 16
Standard Voice Action 30 24
Starter Prompts 2 2
Basic Prompts 2 2
Standard Prompts 4 4
Advanced Prompts 16 16
Help Agent Resolutions 400 0
IT and HR Employee Agent Resolutions 400 0
Customer Engagement Leads Worked 400 320
Speech to Text 150 150
Text to Speech 6,000 6,000
Translation 4,000 4,000

Speech to Text is per hour of transcription, Text to Speech and Translation per million characters. Two unit labels deserve a flag: Translation reads "per 1M characters translated", and the Text to Speech row repeats the word "translated", which reads as a carry-over from the row beneath it rather than a second definition.

Three ratios, then, under one heading. Four rows pay four-fifths in pre-production. Seven pay the same number. Two pay nothing at all. The document does contain a sandbox sentence, and it is the only one of its kind: "For clarity, the Sandbox multiplier applies to pre-production environments, which include sandboxes as well as scratch orgs, when Services are used in such pre-production environments." That sentence settles where the column applies. It says nothing about why a row sits in one band rather than another.

Two reading notes. I read this PDF through page-text extraction rather than a rendered layout: the multipliers are legible, but the table's columns collapse in places, most visibly in the speech rows, and each pair above is read from adjacent cells. And a second voice row, Custom Voice Action, appeared in an earlier extraction of the same file but did not return a production and sandbox pair in the pass I sealed. I have excluded it rather than pair it by inference.

The ratio does not follow the number

Put the two 400-credit action-adjacent rows beside each other. Customer Engagement Leads Worked is 400 in production and 320 in pre-production, a fifth off. Help Agent Resolutions is 400 in production and 0 in pre-production. IT and HR Employee Agent Resolutions is 400 and 0. One production multiplier, two sandbox answers. Whatever the second column encodes, it is not a function of the first.

It is not a function of size either. The largest number in the table, 6,000 credits per million characters of Text to Speech, gets no reduction. Neither does 4,000 for Translation, 150 per hour of transcription, or the four prompt rows. The rows that do move are the ones shaped like an agent doing something to a record, plus one leads row. The two rows shaped like a completed resolution are the only rows that reach zero.

The collapse is the quiet one. In production a Standard Action is 20 credits and an Advanced Prompt is 16, a ratio of five to four. In sandbox both are 16. A customer sizing a build from pre-production consumption would learn that a standard action and an advanced prompt cost the same, which production contradicts.

What one pack divides into

Every figure in this section is my arithmetic on printed values, held apart from every printed number. At the pack price of $500 for 100,000 credits, one credit is $0.005. Against that pack, the table's nine distinct multipliers behave differently: 2, 4, 16, 20, 400 and 4,000 divide 100,000 exactly, giving 50,000 prompts, 25,000 at 4, 6,250 at 16, 5,000 actions at 20, 250 resolutions at 400 and 25 million characters at 4,000. The other three leave a remainder — 3,333 voice actions at 30 with 10 credits left over, 666 hours of transcription at 150 with 100 left over, 16 million characters at 6,000 with 4,000 left over.

None of those remainders strands a credit, because the pricing page states that "Flex credits are fungible across Actions, Prompts, Translations, and Voice Actions" (Agentforce pricing). What the remainders do mean is that no single row closes a 100,000-credit pack on its own arithmetic. The per-unit list figures follow from the same division: 2 credits is $0.01, 4 is $0.02, 16 is $0.08, 20 is $0.10, 30 is $0.15, 150 is $0.75 per hour of transcription, 400 is $2.00, 4,000 is $20.00 per million characters and 6,000 is $30.00 per million characters. The $0.10 per action in the May 2025 launch material is the same division seen from the other end — twenty credits at five-tenths of a cent is ten cents — which is why that figure cross-checks cleanly against the pack (launch release).

Six currencies are printed for the same pack: $500, €500, £400, AU$700, ¥60,000 and 5,000 kronor per 100,000 credits. Divided out — my arithmetic again — that is $0.005, €0.005, £0.004, roughly AU$0.007, ¥0.60 and 0.05 kronor per credit. The pack size is constant across the six. The per-credit amount is not.

Where two printed numbers meet

Four hundred credits at $0.005 is $2.00. The pricing page prints a $ 2 figure on its Help Agent Resolutions line, and at the printed pack price 100,000 credits buys 250 of those at $2.00 each, which is $500. So the page's own resolution figure and the rate card's 400-credit multiplier are the same money at the printed pack price. That is arithmetic on two printed numbers meeting, not a statement that the two nouns name one product.

The same page also says: "Agentforce actions are 20 Flex Credits, while Agentforce Voice actions are 30 Flex Credits." No sentence I read reconciles that general rate with the resolution line. The May 2025 launch release and the May 2025 Help article both print the pack as 100,000 credits for $500 and one action as 20 credits at $0.10 (Help article). Neither prints a sandbox column, neither prints the resolution rows, and neither is dated after the November 2026 rate card.

What the four documents do not say

Read against each other, the rate card, the pricing page, the 2025 release and the 2025 Help article leave these gaps. No reviewed sentence explains why a given row receives four-fifths, parity or zero; only the pre-production definition exists. No reviewed page labels the printed multipliers as list rates, and none prints a customer's contract rate beside them.

The pricing page carries no expiry, rollover or order-end-date rule; the rate card carries one and points at a document the public cannot read: "Flex Credits must be used before the Order End Date set forth in the Usage Details table on the Order Form, and no rollover will be permitted." No reviewed page works an example that carries both multipliers for the same workload, so the pre-production treatment cannot be checked against a printed calculation. And no reviewed page states whether a failed, escalated or reopened agent action is billed once or more than once; the pricing page says only that each action is metered individually.

The rate card does print what happens next: "Usage types, tiers, and associated multipliers may be updated from time to time. Changes to tiers will be treated as multiplier changes." It also flags an alternative unit for some buyers — "For certain customers, the Agentforce Voice Minutes usage type may apply instead" — and it footers itself with technical restrictions it does not enumerate.

What a buyer can still recompute

Twenty standard actions at list cost $2.00, which is one Help Agent Resolution at list. In pre-production those twenty actions cost $1.60 and the resolution costs nothing. The two workloads start level in production and end the column apart, so pre-production consumption cannot be scaled into a production forecast: the production column has to be carried row by row, because the second column moves some rows, holds others and erases two.

The count is a reader's to rebuild. The proportion is the vendor's to name. Salesforce's directory entry records the counterparty this billing model attaches to (Salesforce in the BTW directory).