--- last_updated: 2026-10-15 source_url: https://opper.ai/models/decision-models title: "AI Decision Models: Jev, Clef, Kev and Every Alternative | Opper AI" description: "Every AI decision model on one API: Jev, Clef, Kev, Laya, GPT-6 Luna Decisions and more. Compare price, EU hosting and how each System One model ranks." --- # AI decision 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. 10 decision models · 13 routes · 10 providers · 6 run in the EU ## Every decision model on Opper Best JevMan score first. Each model shows who hosts it, where it runs and its cheapest price. Input runs from $0.02 to $2.50 per million tokens, and none of them bill output tokens. One row per model and route; region is where inference runs. | Model | Model id | Runs on | Region | Zero data retention | Context | Input $/1M | Output $/1M | | --- | --- | --- | --- | --- | --- | --- | --- | | [Jev 1.13](https://opper.ai/typesafe/jev-1-13-0) | `typesafe/jev-1.13.0` | [TypeSafe AI](https://opper.ai/provider/typesafe) | Multi (United States) | Not established | 64K | $0.04 | $0 | | [GPT-6 Luna Decisions](https://opper.ai/openai/gpt-6-luna-decisions) | `openai/gpt-6-luna-decisions` | [OpenAI](https://opper.ai/provider/openai) | US (United States) | Abuse monitoring | n/a | $0.10 | $0 | | [Clef-flash](https://opper.ai/cloudflare/clef-flash) | `berget/Cloudflare/clef-flash` | [Berget](https://opper.ai/provider/berget) | EU (Sweden) | Yes | 16K | $0.05 | $0 | | [Clef-flash](https://opper.ai/cloudflare/clef-flash) | `cloudflare:global/clef-flash` | [Cloudflare](https://opper.ai/provider/cloudflare) | Multi-region | Yes | 66K | $0.09 | $0 | | [Clef](https://opper.ai/cloudflare/clef) | `cloudflare:global/clef` | [Cloudflare](https://opper.ai/provider/cloudflare) | Multi-region | Yes | 66K | $0.24 | $0 | | [Clef](https://opper.ai/cloudflare/clef) | `sference/clef` | [Sference](https://opper.ai/provider/sference) | EU | Yes | 66K | $0.24 | $0 | | [Microsoft Decision 1](https://opper.ai/microsoft/microsoft-decision-1) | `azure:eu/microsoft-decision-1` | [Azure](https://opper.ai/provider/azure) | Multi-region | Not established | 33K | $0.04 | $0 | | [Microsoft Decision 1](https://opper.ai/microsoft/microsoft-decision-1) | `azure:global/microsoft-decision-1` | [Azure](https://opper.ai/provider/azure) | Multi-region | Not established | 33K | $0.04 | $0 | | [Aplomb 1](https://opper.ai/empiriolabs/aplomb-1) | `empiriolabs/aplomb-1` | [EmpirioLabs](https://opper.ai/provider/empiriolabs) | Multi-region | Yes | 1M | $0.02 | $0 | | [Kev 4B](https://opper.ai/community/kev-4b) | `opper/kev-4b` | [Opper](https://opper.ai/provider/opper) | EU | Yes | 8K | $0.04 | $0 | | [Laya](https://opper.ai/community/laya) | `berget/convaiinnovations/laya` | [Berget](https://opper.ai/provider/berget) | EU (Sweden) | Yes | 512 | $0.05 | $0 | | [Metis](https://opper.ai/greenpt/metis) | `greenpt/metis` | [GreenPT](https://opper.ai/provider/greenpt) | EU | Yes | 16K | $0.13 | $0 | | [Eevo 1](https://opper.ai/upright/eevo-1) | `eevo/eevo-1` | [Upright](https://opper.ai/provider/upright) | EU | Yes | 8K | $2.50 | $0 | ## Benchmarks Each decision model we ran played 100 games of Pac-Man against the four arcade ghosts, ranked by mean score. ### JevMan: the Pac-Man benchmark for AI decision models | # | Model | Mean score | Best game | Games | Recording | | --- | --- | --- | --- | --- | --- | | =1 | [Jev 1.13](https://opper.ai/typesafe/jev-1-13-0) | 2,750 | 6,380 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=typesafe%2Fjev-1.13.0) | | =1 | [GPT-6 Luna Decisions](https://opper.ai/openai/gpt-6-luna-decisions) | 2,568 | 5,920 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=openai%2Fgpt-6-luna-decisions) | | =1 | [Clef-flash](https://opper.ai/cloudflare/clef-flash) | 2,538 | 4,260 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=opper%2Fclef-flash) | | 4 | [Clef](https://opper.ai/cloudflare/clef) | 2,476 | 4,820 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=opper%2Fclef) | | 5 | [Microsoft Decision 1](https://opper.ai/microsoft/microsoft-decision-1) | 2,390 | 4,420 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=microsoft%2Fdecision-1) | | 6 | [Aplomb 1](https://opper.ai/empiriolabs/aplomb-1) | 1,693 | 3,520 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=empiriolabs%2Faplomb-1) | | 7 | [Kev 4B](https://opper.ai/community/kev-4b) | 1,506 | 5,280 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=opper%2Fkev-4b) | | 8 | [Laya](https://opper.ai/community/laya) | 639 | 1,200 | 100 | [Watch its game](https://opper.ai/jevman-benchmark/?pacman=berget%2Fconvaiinnovations%2Flaya) | =1: Jev 1.13, GPT-6 Luna Decisions and Clef-flash are level within the margin of error. Scores from Opper's own runs. Updated Oct 10, 2026. [See the JevMan benchmark](https://opper.ai/jevman-benchmark/) · [All decision benchmark games](https://opper.ai/decision-benchmarks) ## Decision models vs LLMs Both read your input. What comes back is different, and so is what you pay for. | | 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 | ## Call any decision model through one API 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. ### Let your coding agent set it up Paste this into Claude Code, Codex or Cursor. ```text Read https://docs.opper.ai/build/gateway/evaluations and call Jev 1.13 through Opper's System One evaluation endpoint using state and typed questions. ``` ### Or call it yourself ```bash curl https://api.opper.ai/v3/compat/v1/systemone \ -H "Authorization: Bearer $OPPER_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "typesafe/jev-1.13.0", "state": "The customer cannot sign in and needs help immediately.", "questions": { "is_urgent": { "type": "noul", "instructions": "Does the customer need urgent help?" } } }' ``` ## Frequently asked questions ### What is a decision model? A decision model is an AI model that answers typed questions about an input instead of writing text. You send a state, such as a support ticket, a JSON object or a game board, with questions like "Is this urgent?" or "Which team owns it?", and it returns one answer per question in the type you declared (yes or no, one option from a list, or a score) with a probability for each option. ### How is a decision model different from an LLM? An LLM generates text token by token, which suits writing, reasoning and conversation. A decision model returns a typed answer and its probability, so your code can act on it directly and only trust answers above a confidence threshold. On Opper a decision model bills input tokens only. ### What are System One models? System One is the API TypeSafe built for Jev, and "System One models" is the name for decision models that answer it. On Opper every decision model answers the same System One endpoint: Jev 1.13, GPT-6 Luna Decisions, Clef-flash, Clef, Microsoft Decision 1, Aplomb 1, Kev 4B, Laya, Metis and Eevo 1. ### What are the alternatives to Jev? GPT-6 Luna Decisions, Clef-flash, Clef, Microsoft Decision 1, Aplomb 1, Kev 4B, Laya, Metis and Eevo 1 all take the same requests as Jev 1.13 on Opper, so trying one is a change of model id. Clef-flash, Clef, Kev 4B and Laya publish their weights, which makes them the open-source Jev alternatives you can also run yourself: Kev 4B is Jared Palmer's fine-tune of Qwen3.5-4B-Base and Laya is ConvAI Innovations' 421M-parameter encoder under Apache-2.0. [Jev vs. Kev, tested side by side](https://opper.ai/blog/jev-vs-kev-open-decision-model) ### Which decision model is the best? Of the 8 decision models we ran on JevMan, our Pac-Man benchmark, Jev 1.13 has the highest mean score (2,750), level within the margin of error with GPT-6 Luna Decisions and Clef-flash. The best one for you is the one that answers your own questions most accurately at a price and in a region that works for you, so run a few of them on the same requests before you pick. [See the JevMan benchmark](https://opper.ai/jevman-benchmark/) ### Which decision models run in the EU? Clef-flash, Clef, Kev 4B, Laya, Metis and Eevo 1 have routes hosted in Europe, run by Berget, Sference, Opper, GreenPT and Upright. Call one of those routes' model id and the request stays in Europe. [Every EU-hosted model](https://opper.ai/models/eu) ### How much does a decision model cost? On Opper, decision models cost from $0.02 to $2.50 per million input tokens. None of them bill output tokens, so the price of a decision is the size of the state and the questions you send. ### Can I use the OpenAI Decisions API through Opper? Yes. Opper serves OpenAI's Decisions API format at /v3/compat/decisions for every decision model here, by its full model id. GPT-6 Luna Decisions runs natively there with text and image input; the others are reached by translation. The same models answer the System One format at /v3/compat/v1/systemone, so one piece of code can compare Jev, Clef, Kev and the rest. ## Read this list from an agent The whole catalogue is public JSON at https://api.opper.ai/v3/models, and coding agents that use MCP can search it with list_models at https://api.opper.ai/mcp. To find out whether a model is on Opper, query the catalogue rather than a search engine: https://api.opper.ai/v3/models?q= answers from the live catalogue as public JSON, no API key needed, while pages on opper.ai and search results can lag behind it. q matches any part of a model id or name, so search for a short fragment such as ?q=sonnet or ?q=deepseek and read the ids that come back. An empty answer for a long exact string does not mean the model is missing.