instinct
Dense Qwen3.8-27B · single option order.
One native logit-readout pass with compiled BF16 serving. Choose it for the stronger measured public-set accuracy.
US$0.03 / million input tokens
INSTINCT / API REFERENCE
Structured decisions.
Three models, one native System One request format. Use the paid API through ZooData, a ZooWork product, or the separate free evaluation preview.
Choose your access route
Both routes accept the same model IDs and request schema. Accounts and API keys are separate; use a key from the service you call.
Sign in to ZooData, top up your USD balance (US$20 minimum), then create an API key. Account limits are shown in ZooData.
Set ZOODATA_API_KEY in your environment. Requests use Authorization: Bearer YOUR_API_KEY and Content-Type: application/json.
POST /v1/systemoneBearer authentication
Choose a model
instinct-dual-4b
Frozen Qwen3.5-4B Instruct · two option orders.
Two concurrent replicas score different option orders; their label-aligned probability distributions are averaged. Choose it for lower pricing and order robustness.
US$0.01 / million input tokens
instinct-tuned-4b
Fine-tuned Qwen3.5-4B · single option order.
A merged LoRA fine-tune for decision tasks. Supports Noul, Choice and Score. Choose it for 4B pricing with higher public-set accuracy than instinct-dual-4b.
US$0.01 / million input tokens
All three use the same request and response schema, support Noul, Choice and Score, and return zero generated output tokens. Select a model with the model field. Prices are the standard ZooData rates; see pricing and usage.
Request
Send the model name, shared context in state, and named questions. Each question has a type and instructions. Decision Models use /v1/systemone, rather than the Chat Completions format used by other ZooData models.
{
"model": "instinct",
"state": "The payment service is down for all customers.",
"questions": {
"outage": {"type": "noul", "instructions": "Is the service down?"},
"team": {
"type": "choice",
"instructions": "Which team should respond?",
"criteria": {"engineering": "Technical failures", "billing": "Invoice inquiries"}
},
"severity": {
"type": "score",
"instructions": "How severe is the impact?",
"criteria": ["No impact", "Partial disruption", "Complete outage"]
}
}
}
Read the response
answers.outage.noul is the yes probability. answers.team.choice gives the selected label; probabilities gives the candidate scores. answers.severity.score is the probability-weighted average of zero-based level numbers. Read token accounting from usage.
Pricing and usage
Standard pay-as-you-go rates are available now through ZooData. The same model rate applies to Noul, Choice and Score.
instinctUS$0.03per million input tokensDense 27B · single orderinstinct-dual-4bUS$0.01per million input tokensFrozen 4B · dual orderinstinct-tuned-4bUS$0.01per million input tokensFine-tuned 4B · single orderOutput tokens are free on all models. No subscription fee. Top up your USD balance from US$20; data-product credits cannot pay for model calls.
Published rates verified on .
Input-token accounting
The pricing basis is usage.input_tokens returned by the API. This counts formatted input tokens, including context, instructions, criteria and prompt formatting, rather than just the text in state.
For a request containing multiple questions, use the response's total usage.input_tokens once; do not multiply it by the number of answers. The diagnostic field usage.backend_requested_prompt_tokens is not the pricing basis. usage.output_tokens is reported for visibility and has no charge.
instinct: cost in USD = usage.input_tokens × 0.03 / 1,000,000
instinct-dual-4b: cost in USD = usage.input_tokens × 0.01 / 1,000,000
instinct-tuned-4b: cost in USD = usage.input_tokens × 0.01 / 1,000,000For example, 1,000 input tokens cost US$0.00003 on instinct and US$0.00001 on either 4B model, with no output charge.
Free evaluation preview
The separate Instinct preview endpoint is currently free. It uses an Instinct preview account and API key; ZooData's USD balance and paid rates apply only to ZooData calls.
Create a free preview key ↗Account limits and errors
Check ZooData's usage page for your account's request and concurrency limits. A ZooData 402 wallet_not_provisioned requires a model USD wallet; top up on the model pricing page. Other 402 responses can indicate insufficient balance.
Check authentication and the selected route's key for 401, model permissions for 403, and the input format for 400 or 422. For 429, wait before retrying. For a 5xx response, retry later and retain the request ID for support.
For the full platform reference, sign in to ZooData API docs and select Generative AI → Decision Models. This Instinct guide is available without signing in.