Pricing
You pay per call. Every price below is what the API itself reported when this page was built (2026-11-22T02:07:45Z); call the API for the price now. A blank means the API did not report it, never zero.
How you pay
- Starter credit: $0.10 per new account (
accounts.free_credit_usdin the API's config, read 2026-11-22T02:07:46Z). Get a key. - Payments: open. You pay from a prepaid key, or in BSV against a signed HTTP 402 quote, or in USDC via x402 (
GET https://api.gaents.com/pay402/healthanswered HTTP 200 at 2026-11-22T02:07:29Z:rails_offered=internal_balance, bsv_tx, x402_usdc). - No charge for empty answers. An answer that delivers nothing is not charged. The API's own words (
GET https://api.gaents.com/v1/models,reasoning_headroom.billing):the charge stays capped at the quote for YOUR max_tokens; reasoning tokens are completion tokens at the listed price, up to that cap; an empty answer is not charged
. The config rule (_comment.billing.charge_empty, minus its type tag;billing.charge_empty=false):false = an upstream 200 that delivered nothing (chat: no text content, tool calls or function call in any choice, streamed or not; seller listing: empty body or JSON null/{}/[]/""; job: 0 rows) is not charged: the hold is released (append-only reversal rows) and the proof says empty_no_charge / the call row empty_released. A provider cost reported for it stays booked as our loss on the route account. true = charge as delivered.
- Each call is quoted first, at a ceiling, and never charged above it. For a model call the ceiling is prompt bytes x prompt price +
max_tokensx completion price (pay402 README); the charge is the call's real cost under the rule below, and the rest of the ceiling is released in the same request. Every charge is a signed row in the public ledger.
The pricing rule
The API's config (services/pay402/config.json, extended by deploy/config.vpsp.json, read 2026-11-22T02:07:46Z) states it like this (_comment.market.model_cost_basis, quoted as written minus its leading ADMIN. type tag):
How model-route calls are priced. provider_list_plus_credit_fee = provider list/reported cost x (1 + market.provider_credit_fee[provider]) x (1 + market.model_spread), never below the real cost. margin_rule = cost / (1 - pricing.rail_fee - pricing.target_gross_margin) (the 60% rule, which still prices jobs and our own Actors' results).
- Model calls
market.model_cost_basis=provider_list_plus_credit_fee(Adam 11-19 decided option b (via coordinator): 'model-route calls are priced at the upstream provider's price plus a spread, not the 60% rule. Apify Actors / our own results stay on the 60% rule.'). Price = the provider's list price x (1 + that provider's credit-purchase fee) x (1 + spread).- Spread
market.model_spread= 0.1 (10%)- Credit-purchase fee
market.provider_credit_fee: 5.5% for the 1 provider with a fee on record, 0% for the rest. So a model price is 1.1605x or 1.1x the provider's list price.- Jobs and our own products
- The 60% gross-margin rule: price = cost / (1 -
pricing.rail_fee-pricing.target_gross_margin) = cost / (1 - 0 - 0.6) = 2.5x cost. Each job's own charged line below says what its cost is measured on. - Seller listings
- The seller sets the price; the platform keeps
market.platform_fee= 5% (Sell).
The per-1M prices below already include the fee and spread: they are the API's our_price_usd_per_*_token x 1,000,000.
Models
From GET https://api.gaents.com/v1/models at 2026-11-22T02:07:29Z. Names are model families as the API lists them. Measured quality per task: Models.
| Model | Input $ per 1M tokens | Output $ per 1M tokens |
|---|---|---|
gaents/claude-haiku |
$0.11605 | $0.58025 |
gaents/claude-opus |
$4.642 | $23.21 |
gaents/claude-sonnet |
$2.321 | $11.605 |
gaents/deepseek |
$0.30173 | $0.44099 |
gaents/deepseek-flash |
$0.0783919 | $0.156784 |
gaents/fable |
$9.284 | $46.42 |
gaents/glm-flash |
$0.046304 | $0.0909832 |
gaents/gpt-oss |
$0.0663134 | $0.265254 |
gaents/grok |
$1.45063 | $2.90125 |
gaents/qwen-27b |
$0.0406175 | $1.74075 |
gaents/qwen-flash |
$0.174075 | $0.545435 |
Worked example: one chat call
A typical call here, taken from the 2613 paid chat calls in the public ledger copy (medians over every payer, our own accounts included): 360 prompt tokens in, 21 completion tokens out. On gaents/glm-flash (the model the start page uses) at the prices above:
360 x $0.046304 / 1,000,000 = $0.0000166694
21 x $0.0909832 / 1,000,000 = $0.00000191065
total = $0.0000185801
The starter credit of $0.10 covers 5382 such calls. The real charge is the provider-reported cost of your call under the rule above, so it can differ a little from this estimate; it is never above the quote.
Jobs
From GET https://api.gaents.com/v1/jobs answered HTTP 200 at 2026-11-22T02:07:45Z. Prices in USD as the API reports them; "charged" is the API's own sentence.
| Job | Unit | Price | How it is charged |
|---|---|---|---|
| Transcribe an audio or video file (Whisper small) | audio-minute | $0.06 per audio-minute | Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| Transcribe an audio or video file (Whisper large-v3-turbo) | audio-minute-premium | $0.1003 per audio-minute-premium | Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| Render a website's pages in a real browser | page-rendered | $0.013 per page-rendered | Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| YouTube video transcript (captions) | transcript | $0.0033 per transcript | Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| Google Trends interest over time for one keyword (worldwide, 12 months) | keyword-result | $0.0015 per keyword-result | Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| A YouTube channel's latest videos | result | $0.0002 per result + $0.0015 per run |
Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| Top comments on a YouTube video | comment | $0.0001 per comment + $0.0008 per run |
Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| YouTube search results for one query | result | $0.0001 per result + $0.0009 per run |
Actor's own event counts x price, only when a usable row is delivered; never above the quote |
| OCR an image to text (Tesseract) | call | up to $0.0000313398 per call | measured CPU seconds of this call x our compute cost, by the 60% rule; never above the quote; no charge for an empty result or a failed run |
| Convert a document to Markdown (MarkItDown) | call | up to $0.0000626797 per call | measured CPU seconds of this call x our compute cost, by the 60% rule; never above the quote; no charge for an empty result or a failed run |
| Extract named entities of the types you name (GLiNER, zero-shot) not live yet | call | no price reported | measured CPU seconds of this call x our compute cost, by the 60% rule; never above the quote; no charge for an empty result or a failed run |
| Text to speech (Piper, English voices) | call | up to $0.000125359 per call | measured CPU seconds of this call x our compute cost, by the 60% rule; never above the quote; no charge for an empty result or a failed run |