Browse AI models that answer typed Choice, Score and yes/no questions with calibrated probabilities, for routing, classification and moderation through the API.
POST /api/v1/decisions. They cannot be used in chat.Example request body
{
"model": "jaredpalmer/kev-9b",
"state": "Customer message: I was charged twice for the same order.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this message?",
"criteria": {
"billing": "Payments, refunds or invoices",
"technical": "Bugs or errors in the product"
}
}
}
}Jared Palmer's Kev 9B, hosted by Gradiated, answers typed Choice, Score and Noul questions about text or JSON state without generating text. Text input only, with an 8,192-token context. Use the Decisions or System One API, or a questions response format through Chat Completions, Messages or Responses. Input costs $0.03 per million tokens, cached input $0.006 per million, and output is free.
Surogate's Rune 26B-A4B is a text-only decision model for classification, scoring and yes/no judgments. It returns Choice, Score and Noul answers with calibrated probabilities, with up to 256 questions per request, 208 choice options and 2–10 score levels. Use /v1/decisions or /v1/systemone; typed questions are also supported through Chat Completions, Messages and Responses. The model has a 262,144-token context window, subject to local request admission limits. It does not generate conversational text or reasoning, and does not support tools, images or streaming. Only input tokens are billed, with a lower rate for cached input.
Subquadratic's long-context decision model returns typed Choice, Score and Noul answers with probabilities. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Accepts text, JSON and up to four inline images, each at most 512 by 512 pixels. Included in subscriptions at the regular input-token rate. Only input tokens are billed; output is free.
vLLM Semantic Router’s d3 is the current 27B decision model for complex classification and policy judgments. Runs in an attested TEE (Trusted Execution Environment), with encrypted requests and attestation verified by NanoGPT. It evaluates text against predicate, choice and score questions and returns calibrated probabilities without generating text. Supports a 32,768-token context, up to 64 questions per request, and up to 26 choice options and 10 score levels. Text input only; reasoning, tools and streaming are unsupported. Use the native input and questions-array format through the Decisions API, or supported text questions through System One compatibility. The public ID decision-large is a mutable size alias, not an immutable checkpoint.
vLLM Semantic Router’s d3 Flash is the current 9B decision model for triage, routing and policy evaluation. Runs in an attested TEE (Trusted Execution Environment), with encrypted requests and attestation verified by NanoGPT. It evaluates text against predicate, choice and score questions and returns calibrated probabilities without generating text. Supports a 32,768-token context, up to 64 questions per request, and up to 26 choice options and 10 score levels. Text input only; reasoning, tools and streaming are unsupported. Use the native input and questions-array format through the Decisions API, or supported text questions through System One compatibility. The public ID decision-medium is a mutable size alias, not an immutable checkpoint.
vLLM Semantic Router’s d3 Mini is the current 4B decision model for fast first-pass classification and high-volume routing. Runs in an attested TEE (Trusted Execution Environment), with encrypted requests and attestation verified by NanoGPT. It evaluates text against predicate, choice and score questions and returns calibrated probabilities without generating text. Supports a 32,768-token context, up to 64 questions per request, and up to 26 choice options and 10 score levels. Text input only; reasoning, tools and streaming are unsupported. Use the native input and questions-array format through the Decisions API, or supported text questions through System One compatibility. The public ID decision-small is a mutable size alias, not an immutable checkpoint.
Perplexity's 27B decision model, succeeding PPLX Decider v1.1 with the same API contract; requests for perplexity/pplx-decider-v1.1-27b now run on v1.2. Typed Choice, Score and Noul decisions. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Accepts inline images through the Perplexity endpoint. The local image limit is four images, each at most 512 by 512 pixels. Input tokens are billed; output is free.
Bevel's Winnow-12B fine-tune for Choice, Noul and Score decisions, with inference in Germany. Use /v1/decisions or /v1/systemone, or the typed questions format through Chat Completions, Messages and Responses. Supports text and JSON, eight questions per request, up to 64 Choice options, and 2–10 Score levels. Output is free. The latest ID follows new releases; version 1.0.0 is pinned. USD pricing uses a fixed conversion of the provider's EUR tariff. Personal data requires a DPA with the provider.
Bevel's Winnow-12B fine-tune for Choice, Noul and Score decisions, with inference in Germany. Use /v1/decisions or /v1/systemone, or the typed questions format through Chat Completions, Messages and Responses. Supports text and JSON, eight questions per request, up to 64 Choice options, and 2–10 Score levels. Output is free. The latest ID follows new releases; version 1.0.0 is pinned. USD pricing uses a fixed conversion of the provider's EUR tariff. Personal data requires a DPA with the provider.
TypeSafe's Jev 1.14 (jev-1.14.0) returns typed decisions for routing, classification, scoring, and yes/no judgments, with improved multilingual performance. Use /v1/decisions, TypeSafe SDK compatibility through /api/v1/systemone, or a questions response format through Chat Completions, Messages, or Responses. Text and JSON input only on this route, with 64k tokens per request and 32k per question. Experimental images, Box localization, and extended context are not enabled.
Typed Choice, Score and Noul decisions about text, inline images and one PCM16 WAV audio clip up to 10 seconds. Use /v1/decisions or /v1/systemone. Up to four 512x512 images; image and audio bytes share a 512 KiB base64 / 384 KiB decoded limit. Text admission is limited to 1,800 estimated tokens including questions. Video is not enabled. No ZDR, audio moderation or media redaction. Compatibility APIs accept text only. Actual input tokens are billed; output is free.
Redpill lists Kev 4B as running on Phala in a TDX-attested TEE (Trusted Execution Environment). NanoGPT does not currently enforce attestation or verify signed receipts for this route. Jared Palmer's Kev 4B scores typed Choice, Score and Noul questions about text or JSON state in one forward pass, returning decisions and probabilities without generating text. Based on Qwen3.5-4B-Base with Apache 2.0 weights, it supports English and an 8,192-token context. Use /v1/decisions or /api/v1/systemone, or a questions response format through Chat Completions, Messages or Responses. Choice and Score support 1–255 options or levels. Input, including cached input, costs $0.042 per million tokens; output usage is free.
Subquadratic's long-context decision model returns typed Choice, Score and Noul answers with probabilities. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Accepts text, JSON and up to four inline images, each at most 512 by 512 pixels. Only input tokens are billed; output is free.
Typed Choice, Score and Noul decisions. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Accepts inline images. Text admission is limited to 1,800 estimated tokens to avoid truncation; up to four images, each at most 512 by 512 pixels. Input tokens are billed; output is free.
Typed Choice, Score and Noul decisions. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Accepts inline images. Text admission is limited to 1,800 estimated tokens to avoid truncation; up to four images, each at most 512 by 512 pixels. Input tokens are billed; output is free.
Typed Choice, Score and Noul decisions about text, JSON or inline images for classification, routing and scoring. Use /v1/decisions or /v1/systemone; text-only questions compatibility is available through Chat Completions, Messages and Responses. Uses OpenAI’s native Decisions API. Input tokens are billed; output is free.
Liquid AI's D1 evaluates application state and returns typed Choice, Score and yes/no answers with calibrated probabilities. Use /v1/decisions or /v1/systemone, or a questions response format through Chat Completions, Messages or Responses. It supports a 65,536-token context and generates no text tokens; only input tokens are billed.
Mapika's Decider, quantized and hosted by LLM Tech, answers typed Choice, Score and Noul questions about application state. Use /v1/decisions or /v1/systemone, or a questions response format through Chat Completions, Messages or Responses. It returns probabilities and decisions without generating text; only input tokens are billed.
Mapika's Decider, quantized and hosted by LLM Tech, answers typed Choice, Score and Noul questions about application state. Use /v1/decisions or /v1/systemone, or a questions response format through Chat Completions, Messages or Responses. It returns probabilities and decisions without generating text; only input tokens are billed.
Mapika's Decider, quantized and hosted by LLM Tech, answers typed Choice, Score and Noul questions about application state. Use /v1/decisions or /v1/systemone, or a questions response format through Chat Completions, Messages or Responses. It returns probabilities and decisions without generating text; only input tokens are billed.
TypeSafe's Jev 1.13 is a System One decision model for routing, classification, scoring, and other application decision points. Use /v1/decisions directly, point the official TypeSafe SDK at the /api base URL for /api/v1/systemone compatibility, or pass a questions response format through Chat Completions, Messages, or Responses; Jev returns typed answers with calibrated probabilities instead of generated text.
An alias that routes to the latest TypeSafe Jev System One decision model. Use /v1/decisions directly, point the official TypeSafe SDK at the /api base URL for /api/v1/systemone compatibility, or pass a questions response format through Chat Completions, Messages, or Responses; Jev returns typed answers with calibrated probabilities instead of generated text.
