One workflow. One meter. Find it.

Which meter did that hit?

Grok, X, xAI, voice, and Cursor look like one connected machine. Billing does not work that way. Follow the workflow, not the model name.

The interface chooses the meter.Using Grok in two products does not make them share a balance.

Built-in X search is not your public X API.Grok app search is metered with the app; xAI API X Search uses API credits.

Cursor has two distinct Grok lanes.Linked Grok Bot uses a weekly grant; Grok model calls use Cursor's model pool. Neither reaches into xAI API credits.

Six provider cards

Same model name, different checkout

Each card answers three things: where usage is metered, where the subscription wall sits, and what changed in September 2026.

xAI

Mostly verified

xAI

Meter: Grok app tasks use the consumer pool. API models, tools, and API voice use the console.x.ai API billing account: prepaid credits first, then optional invoicing if enabled. No-code Voice Agent Builder starts at $0.05/min plus telephony, but its exact billing account is not documented.
Boundary: SuperGrok does not pay for xAI API use.
Sep ’26: On Sep 21 at noon PT, API X Search becomes $5/1k posts and $10/1k profiles fetched.

Official pricing ↗

Separate rail

Public X API

Meter: Direct api.x.com calls use the X developer account’s own pay-per-use bill and quota.
Boundary: It is separate from Grok consumer search and xAI’s server-side X Search tool.
Sep ’26: No September pricing change was found for the public X API; the confirmed change belongs to xAI API X Search.

Public API context ↗

CUR

Official docs live

Cursor

Meter: Linked Grok Bot uses a separate weekly grant. Grok model calls use Cursor’s first-party pool. Claude, GPT, and Gemini use Cursor’s external model pool. Opt-in overage is billed at API rates with no markup.
Boundary: Cursor usage does not draw from your xAI API balance.
Sep ’26: Current docs call the SuperGrok link a usage grant, not an Ultra plan; third-party “free Ultra” claims conflict.

Cursor’s linked-grant docs ↗

Official boundary

Google

Meter: Gemini API uses its own prepaid or Cloud billing: model tokens plus tool lines. Search and Maps grounding each include 5,000 prompts/month, then cost $14/1k queries. Vertex AI runs on Google Cloud billing.
Boundary: Google AI plans cover AI Studio UI benefits; API keys and external apps bill separately. AI Studio agents such as Deep Research require a paid API key.
Sep ’26: Gemini 3.8 Live rates and Flash introductory pricing were reported by secondary sources, but were not confirmed on official docs. Treat both as unconfirmed.

Google AI plan boundary ↗

Official docs live

Anthropic

Meter: Claude API is token-billed. Sonnet 5 is $2/$10 per million input/output tokens; Opus 5 is $5/$25; Fable and Mythos 5.1 are $10/$50. Batch is 50% off.
Boundary: Subscriptions and Console/API are separate, except paid plans can opt into a small Agent SDK credit pool at API rates: Pro $20, Max 5x $100, Max 20x $200 per cycle.
Sep ’26: Fable/Mythos 5.1 cache reads are 0.025x, or $0.25/MTok. The launch-date and “up to 45% cheaper” claims are secondary.

Claude API pricing ↗

OAI

Official pricing live

OpenAI

Meter: API use is tokens plus tools: web search $10/1k calls, file search $2.50/1k calls plus storage, and containers from $0.03 per 20-minute 1 GB session. Research found no separate Agents API platform fee, but that point came from a secondary source.
Boundary: ChatGPT plans and API billing are separate. Plus/Pro can buy extra Codex and Sora credits after included limits.
Sep ’26: GPT-6 Astra is $10/$50 per million input/output tokens. GPT-5.6 Sol is verified at promotional $4/$20 through at least Nov 21; the promo start date is unconfirmed. A claimed Sep 24 Sora sunset is also unconfirmed.

OpenAI API pricing ↗

What about Glint checking email or doing research?

This research did not audit Muse billing, so it would be dishonest to assign that work to one of the meters above. Nothing found here shows that reading Gmail through Muse automatically spends Grok, xAI API, Cursor, Google, Anthropic, OpenAI, or public X API balances. A named handoff to one of those products starts that product’s meter; Muse’s own meter remains a separate open question.

Interactive router

Pick what you did

The active meter lights up. Switch workflows to see where the cost or quota lands.

Consumer app

Scheduled Grok task that searches X or the web

The task runs inside Grok, so it draws from the consumer task/compute pool tied to SuperGrok. It does not use your xAI API prepaid credits, and it does not count against your public X API quota.

GROK APP

usage flows here

SUPERGROK POOL

Charged toSuperGrok task / compute pool

Not charged toxAI API · public X API

WatchConsumer quota

The easy-to-miss correction

Two “X Search” products. Two different bills.

console.x.ai / docs.x.ai

xAI API server-side X Search

The model calls the x_search tool. This spends xAI API credits. On September 21, 2026, pricing changes from per-call to per-item fetched.

From 21 Sep · $5 / 1k posts · $10 / 1k profiles

api.x.com

Public X API

Your code calls X's developer API directly. It has its own account, pricing, quota, and invoice. The September 21 xAI change does not apply here.

Separate product · separate meter

≠

Voice is two products too

API voice is clear. Builder billing is not.

Verified meter

Speech-to-speech API

$0.08 / minute

API consumption uses the same console.x.ai billing account as text calls and server-side tools: prepaid credits first, then optional invoicing if enabled. With the default $0 invoiced limit, requests stop when credits are depleted. Audio is $4.80 per hour, plus $0.004 text input.

Billing pool unverified

No-code Voice Agent Builder

$0.05 / minute, starting

Telephony on xAI phone numbers adds charges. The available sources did not establish whether Builder usage draws from the console.x.ai prepaid pool or a separate account.

Do not collapse these into one answer.

Your API voice calls clearly use API credits. Your no-code phone agents have published starting rates, but their exact billing-account relationship remains an open question.

Cursor reality check

The “free Ultra” story conflicts with the live docs.

Linking SuperGrok gives Cursor usage

Yes. It grants a weekly Grok Bot allowance while the linked subscription is active.

Confirmed

Linking swaps the account to Ultra

No. Cursor calls it a usage grant, not a plan, and says it does not swap you to Ultra.

Contradicted

Grok in Cursor uses xAI API credits

No. Grok is treated as a first-party Cursor model and draws from Cursor's pool.

Extra Cursor usage is automatic

No. On-demand must be enabled. If enabled, overage is billed at API rates with no markup.

Opt-in

Sources & confidence

Official documentation leads where available. Consumer-plan and no-code Builder boundaries still rely partly on third-party reporting, so those caveats stay visible.

xAI API pricing ↗ xAI X Search tool ↗ xAI credits and billing ↗ Cursor linked SuperGrok grant ↗ Cursor on-demand overages ↗ xAI Voice Agent Builder announcement ↗ Gemini API pricing ↗ Google AI plan boundary ↗ Claude API pricing ↗ Claude subscription/API boundary ↗ Claude Agent SDK credits ↗ OpenAI API pricing ↗ ChatGPT/API billing boundary ↗