System One models
A decision model answers typed questions about your input (yes or no, one option from a list, or a score) with a probability for every answer instead of written text. Compare every decision model Opper serves, from TypeSafe's Jev to Clef, Kev and the other open-source alternatives, and call any of them with one API key.
Best JevMan score first. Each model shows who hosts it, where it runs and its cheapest price. Input runs from $0.01 to $0.23 per million tokens, and none of them bill output tokens.
Jev 1.14 is TypeSafe AI's System One model for decisions, classification and scoring, with improved multilingual support and experimental image and bounding-box questions.
d1-ultra, Liquid AI's most capable decision model, answers yes/no, choice and score questions about text or JSON with a calibrated probability for every option.
Evaluates up to 64 typed questions about text or JSON and returns answer probabilities.
Perplexity Decider 1.2 is a 27B decision model that classifies and scores text, JSON and images with typed probabilities over a 256K-token context.
d1.1, Liquid AI's default decision model, answers yes/no, choice and score questions about text or JSON with a calibrated probability for every option.
Jev 1.13 is TypeSafe AI's System One model: it answers true/false, choice and rubric-score questions with calibrated probabilities instead of writing text.
GPT-6 Luna Decisions is OpenAI's gpt-6-luna on its Decisions API: it answers predicate, choice and score questions about text and images with probabilities instead of writing text.
Clef-flash, Cloudflare's fast 9B open-weight decision model, answers yes/no, choice and score questions about text or JSON with a probability for every option in tens of milliseconds.
Clef, Cloudflare's 27B open-weight decision model, answers yes/no, choice and score questions about text or JSON with a probability for every option instead of writing text.
Fast decisions over text with calibrated probabilities for predefined answer options.
Perplexity Decider Flash 1.2 is a 4B decision model for classifying and scoring text, JSON and images with typed probabilities over a 256K-token context.
Aplomb 1, EmpirioLabs' first model: a decision model that answers yes or no, choice and score questions about text, JSON, images, audio and video with calibrated probabilities, over a 1M-token state.
Laya, ConvAI Innovations' open encoder decision model, answers yes/no, choice and score questions about short text with calibrated probabilities in a single forward pass.
Joint decisions over shared state: yes/no, choice and score questions, with probabilities for every answer.
Eevo-1, Upright's consequential decision model, answers yes/no, choice and score questions about a case with a probability for every option, weighing what each option leads to for the people affected.
Clef-omni, Cloudflare's 30B mixture-of-experts decision model with 3B active parameters, answers yes/no, choice and score questions about text, JSON, images, audio or video with a probability for every option instead of writing text.
Each decision model we ran played 100 games of Pac-Man against the four arcade ghosts, ranked by mean score.
=1: Jev 1.14, Eevo-1.1 and reflex are level within the margin of error.
Both read your input. What comes back is different, and so is what you pay for.
| Compared | Decision model | LLM |
|---|---|---|
| You send | A state (a ticket, a JSON object, a game board) and typed questions about it | A prompt |
| You get back | One answer per question from the type you declared: yes or no (noul), one option from a list (choice) or a score, each with its probability | Generated text, or JSON when you ask for it |
| Your code | Acts on the answer and sets a confidence threshold, with nothing to parse | Parses and validates the text |
| Billed on Opper | Input tokens only | Input and output tokens |
| Built for | Triage, routing, moderation, guardrails, verification and game moves | Writing, reasoning, tool use and conversation |
Every decision model here answers the System One API at /v3/compat/v1/systemone and OpenAI's Decisions API format at /v3/compat/decisions, so moving from Jev to Kev or Clef is a change of model id. GPT-6 Luna Decisions runs natively on the Decisions format, with image input; the others are reached by translation.
Paste this into Claude Code, Codex or Cursor.
curl https://api.opper.ai/v3/compat/v1/systemone \-H "Authorization: Bearer $OPPER_API_KEY" \-H "Content-Type: application/json" \-d '{"model": "typesafe/jev-1.14.0","state": "The customer cannot sign in and needs help immediately.","questions": {"is_urgent": {"type": "noul","instructions": "Does the customer need urgent help?"}}}'
This list as data. The same models, prices and regions as markdown at opper.ai/models/decision-models.md.
The whole catalogue. The model catalogue is public JSON at https://api.opper.ai/v3/models, no API key needed. Filter it with query parameters, for example ?type=llm&limit=3, and page through it with limit and offset. Use it when you can fetch a URL but cannot connect to an MCP server. Coding agents that use MCP can connect to https://api.opper.ai/mcp. Two of the server's tools need no account and no sign-in: list_models searches the model catalogue by name, type, provider or capability, and get_guide returns short setup guides. Everything that touches your account needs you to approve the connection in the browser first. About the MCP server
