Pricing
Pay for answers, not attempts.
A call costs credits only when it delivers one: a markdown read, or an extraction that cleared the quality bar. Failures and low scores cost nothing, and every response says what it cost.
Free during beta · nothing is charged today
Credits
What one call costs
The weights track what the answer cost to produce. They are computed by the pipeline itself and returned as metadata.credits.
| Call | Credits |
|---|---|
| Markdown read (contrie_read, format "markdown") | 1 |
| Structured data from the page itself (JSON-LD, no model) | 1 |
| Extraction answered by a free or budget model | 1 |
| Extraction that escalated to a standard, vision or premium model | 3 |
| Extraction that needed the frontier model | 10 |
| Result under 70, RENDER_REQUIRED, fetch failure, models unavailable | 0 |
Plans
Three plans, one ladder
Every plan gets every feature. Plans differ in credits and in how far up the model ladder an extraction may climb.
Free
currentEvery feature, free-tier models
$0 forever
- 1,000 credits a month
- Free-tier models (hedged for speed)
- Markdown reads, JSON Schema, streaming, hosted MCP
- 10 requests a minute
Builder
at launchFor agents in production
$19 /month
- 10,000 credits a month
- Budget and standard tiers (DeepSeek V4, GPT-5.6 Luna, Haiku 4.5)
- Escalation ladder up to the standard tier
- 30 requests a minute
Pro
at launchThe full ladder, frontier included
$79 /month
- 100,000 credits a month
- Full fleet: vision cascade, premium and frontier escalation
- Claude Sonnet 5 when the score demands it
- 100 requests a minute
Pay as you go, for agents that only want a key: $1 per 1,000 credits at launch. Prices are set; billing is not on yet.
Questions
- What exactly costs a credit?
- An answer. A markdown read is 1 credit. An extraction that clears the quality threshold of 70 costs 1, 3 or 10 credits depending on the model tier that produced it — the response tells you which, in metadata.credits. Anything under 70, a RENDER_REQUIRED refusal, a fetch failure or a rate-limited fleet costs 0.
- Why tier the price by model instead of charging per page?
- Because that is what the answer cost us. Structured data on the page and free-tier models cost nothing to run; standard and premium models cost real money; the frontier model costs an order of magnitude more. Pricing by tier keeps the cheap path cheap instead of averaging everyone up.
- What is charged today?
- Nothing. Every account is on the Free plan during the beta. Credits are counted so you can see what a workload would cost, and the quota resets on the 1st of each month. Paid plans and pay-as-you-go open when billing launches; existing keys keep working.
- Do I pay for retries the pipeline makes on its own?
- No. Escalation is internal. If the first model scores 60 and the second scores 92, you pay for one answer at the second model's tier. The trace shows every attempt either way.
- What about the rate limit?
- Rate limits protect the fleet and apply per key regardless of credits: 10 requests a minute on Free, more on paid plans. A rate-limited call costs 0 credits.