Billing & usage

Credits are the independent quota and billing unit agents spend when they call Beatra skills. There is no fixed credit-to-currency exchange rate. Top-up offers define how many credits a payment buys in each market; public model prices define how many credits an operation consumes.

How charging works

Generation, speech synthesis, voice, music, and dedicated image/video understanding tasks are prepaid. Beatra charges the price of the accepted task at submit; generation whose final quantity varies settles against measured usage on success. Understanding uses the full number of accepted images, or input-video seconds rounded up, so the amount charged at submit is exact. Public social-data reads (beatra.social.execute) are also prepaid: the credit price is 6, 12, 30, or 60, and it is confirmed before the task is created.

If the balance cannot cover the prepaid amount, the API returns 402 insufficient_balance and no task is created.

The AI Director (beatra.speech.prepare_script, POST /v1/audio/speech/scripts) is postpaid: it is charged after it succeeds, from the input, output and reasoning tokens it actually used. To start, the balance must cover its estimate for the text: the capability catalog's estimate per 1,000 characters, prorated to the length of the text (counting at least 1,000 characters), plus the estimate of any other AI Director task in progress on the account. Otherwise the API returns 402 insufficient_balance with details.reason: director_estimate, and no task is created. The charge is the actual usage, usually below the estimate; if it is more than the balance at that moment, the charge stops at the balance.

How many at a time

When this many creations are already in progress, a new request is not started and nothing is charged. Wait for one to finish and retry the same request, or pick a plan that allows more at a time. REST per-minute limits are documented on the rate-limits page.

ElevenLabs speech has a separate cap: at most two ElevenLabs speech tasks in progress per account, whatever the plan. Free clone previews count toward it; MiniMax and Doubao tasks do not. A third request returns 429 rate_limited with details.reason: speech_inflight_limit and nothing is charged.

The AI Director has its own cap as well: at most two AI Director tasks in progress per account, whatever the plan, and a limited number across the platform. A third request returns 429 rate_limited with details.reason: director_inflight_limit (over MCP, the plan limit above can answer first); while the AI Director is at capacity, a request returns 429 with details.reason: director_busy. Neither starts a task or charges anything. For director_inflight_limit, wait for one of the account's AI Director tasks to finish, then submit again; for director_busy, retry after details.retry_after_seconds (over REST, the Retry-After header carries the same seconds).

Prepaid understanding

• POST /v1/images/responses charges the accepted image_count at 5 credits per image, up to eight images.
• POST /v1/videos/responses charges the accepted input_video_seconds at 0.4 credits per second; milliseconds are rounded up to a whole second.
• detail and provider token usage do not change the price.
• Failed or confirmed-canceled understanding tasks automatically refund the amount charged at submit through the standard prepaid refund path.

Successful prepaid usage settlement

For a task whose final billable quantity depends on accepted media or measured output, the submission debit is an estimate. On success, Beatra settles against authoritative measured usage using the price snapshot captured at submission:

• lower actual usage returns the unused estimate in refunded_credits;
• equal usage needs no wallet adjustment;
• higher valid actual usage within the accepted request contract adds a supplemental debit to charged_credits.

Measured usage that is inconsistent with its own input/output breakdown or exceeds the accepted request and model limits is never applied as actual usage or as a supplemental debit. The original submission debit remains unchanged while the task stays pending settlement for operator review.

The final amount is returned directly as billing.net_charged_credits (and is equal to billing.charged_credits - billing.refunded_credits). A successful overage can leave the wallet below zero; the artifact remains valid, while new billable work is rejected until the balance is topped up.

Automatic refunds on failure

When a task is failed or cancellation is confirmed and the task becomes canceled, the held credits are automatically refunded to source — no request needed. A cancellation conflict means the task is still running and does not trigger a refund:

"billing": {
  "charged_credits": 48.0,
  "refunded_credits": 48.0,
  "net_charged_credits": 0.0
}

Billing fields in responses

Every task response carries:

Balance and top-up

Credit purchases are one-time payments in CNY with WeChat Pay. View your balance and add credits at console.beatra.cn/wallet?intent=buy.

Credits belong to the account you are signed in to when you add them, and an Agent can only spend the credits of the account it is connected to. The top-up page names that account ("Adding credits to: …"). If you have two accounts, see Accounts and merging.