OpenAI compatible API · Attested · Public status

Structured Decisions

Compare every TrustedRouter route for Structured Decisions, including token pricing, context limits, privacy policy, regional availability, measured uptime, and API support.

Verify gateway

Onebase URL to migrate

100sof models and routes

0prompt or output logs. Always.

`neurometric/structured-decisions`

US providers Neurometric AI

All models

Publisher	Neurometric AIneurometric
AI IQ	No AI IQ profile matched yet.
Context	65,536 tokens
Providers	Neurometric AI
Credits routes	1 provider route
Modes	chat US providers
Billing	Credits

How to use Structured Decisions

Structured Decisions answers one bounded decision about the facts you supply, choosing from labels you define: yes/no, one of several options, or an ordered level. Instead of generating text, it reads the answer from the model's probabilities over your labels, so the result is always exactly one of them, even if the prompt asks for something else. Put rubrics or policies in a system message or in the question. Good for routing, triage, eligibility checks and policy decisions.

Input	OpenAI chat completions. Give the labels in one of two ways: 1) A user message whose content is a JSON object: {"state": <facts: string or object>, "question": <string>, "options": [labels] or {label: description}}. Use "levels": [ordered labels] instead of "options" for a scale (e.g. ["low","medium","high"]). 2) response_format {"type":"json_schema"} whose schema has an "answer" property with an "enum" of labels; the whole conversation is used as the facts. System messages (rubric, policy, definitions) are included with the facts. Up to about 120 labels per decision. Sampling parameters (temperature, top_p, seed) are accepted and ignored. A request with no labels is answered by text generation instead.
Output	choices[0].message.content is exactly {"answer":"<one of your labels>"}. A top-level "decision" object adds: "type" ("choice" or "score"), "answer", "confidence" (0-1, how decisively the model applied your question as worded; not a calibrated probability), and "probabilities" (label -> probability). For levels, "score" is the expected level index (0 = first level). usage.completion_tokens is 0: only input tokens are billed. Streaming returns the same answer in one content chunk.

Input

OpenAI chat completions. Give the labels in one of two ways: 1) A user message whose content is a JSON object: {"state": <facts: string or object>, "question": <string>, "options": [labels] or {label: description}}. Use "levels": [ordered labels] instead of "options" for a scale (e.g. ["low","medium","high"]). 2) response_format {"type":"json_schema"} whose schema has an "answer" property with an "enum" of labels; the whole conversation is used as the facts. System messages (rubric, policy, definitions) are included with the facts. Up to about 120 labels per decision. Sampling parameters (temperature, top_p, seed) are accepted and ignored. A request with no labels is answered by text generation instead.

Output

choices[0].message.content is exactly {"answer":"<one of your labels>"}. A top-level "decision" object adds: "type" ("choice" or "score"), "answer", "confidence" (0-1, how decisively the model applied your question as worded; not a calibrated probability), and "probabilities" (label -> probability). For levels, "score" is the expected level index (0 = first level). usage.completion_tokens is 0: only input tokens are billed. Streaming returns the same answer in one content chunk.

Example input

{"model": "neurometric/structured-decisions", "messages": [{"role": "user", "content": "{\"state\": \"Customer: I was charged twice for my March invoice and need one charge reversed.\", \"question\": \"Which team should handle this ticket?\", \"options\": {\"billing\": \"Charges, refunds, invoices\", \"technical\": \"Bugs, outages, integrations\", \"sales\": \"Pricing, plans, upgrades\"}}"}]}

Example output

{"choices": [{"message": {"role": "assistant", "content": "{\"answer\":\"billing\"}"}}], "decision": {"type": "choice", "answer": "billing", "confidence": 0.984, "probabilities": {"billing": 0.997, "technical": 0.003, "sales": 0.0}}, "usage": {"prompt_tokens": 179, "completion_tokens": 0, "total_tokens": 179}}

Current route evidence

Live leaderboard

$0.03165/1MLowest Credits input price

$0.01/1MLowest Credits output price

not enough dataFastest measured p50 TTFT

not enough dataHighest measured throughput

Recent provider-route uptime spans not enough data across 1 route and 1 provider. Price comes from the current Credits catalog. Speed and uptime come from routed probes rather than vendor claims.

Providers serving this model

Sorted by lowest combined prompt and completion price. Use "provider": {"sort": "throughput"} in the request body (or append :nitro to the model id) to route to the fastest provider instead.

Trust boundary: TR router attestation verifies the TrustedRouter gateway only. Provider policy separately describes the upstream model host's retention, confidential compute, and E2EE posture.

Provider	Input	Cached input	Output	Weight format	TR router attested	Provider policy
Neurometric AIneurometric	$0.03165/1M	Not published	$0.01/1M	Unknown	yes	no verified privacy claim

Compare Structured Decisions

All comparisons

Structured Decisions vs Qwen: Qwen3.8 27B18 combined provider routes

Questions

What model ID should I use for Structured Decisions?

Use neurometric/structured-decisions as the model field with the TrustedRouter OpenAI-compatible API. Credits routing and provider fallback happen behind that model ID.

Which providers serve Structured Decisions?

TrustedRouter currently lists Neurometric AI for Structured Decisions. Provider availability and routing eligibility can change as catalog and health data update.

How much does Structured Decisions cost through TrustedRouter?

The current lowest Credits input price is $0.03165/1M and the lowest Credits output price is $0.01/1M. Prices are per one million tokens and come from the current route catalog.

How do I require zero data retention for Structured Decisions?

Set provider.min_privacy to zdr on the request. TrustedRouter considers only routes with a recorded zero-data-retention posture and fails closed if no eligible route remains.