Repository navigation
open source alternative to microsoft decision 1
jevos is an open-source (MIT) decision model that does the same kind of job as Microsoft-Decision-1, but on your own CPU, with the weights published for anyone to download. Both read a text, take yes/no, multiple-choice or score questions with a fixed set of allowed answers, and return a probability per answer instead of generated text. Microsoft-Decision-1, announced by Microsoft on 2026-10-09, is reached through a deployment in Microsoft Foundry or through OpenRouter and billed per input token; jevos is one binary and a model folder that you run yourself. We know of no measurement of the two against each other, so this page compares what each one is, not which is faster or more accurate.
We build jevos. We are not affiliated with Microsoft.
The real choice is where the text goes. A hosted model sends every text to someone else's data center and adds a network round trip to every decision; a local one keeps the text on the machine and works offline. Those properties hold whatever a benchmark says.
This page is what Microsoft says its model is, how jevos lines up with it, what running locally changes, what jevos does not have, and how to choose.
Microsoft calls it "our new model for fast decision-scoring". This section is Microsoft's own description:
- What it returns. "When given a fixed set of answer options, Microsoft-Decision-1 provides a calibrated probability score for each option." It "doesn't generate a free-form response or written rationale."
-
Question types.
noul(a probability from 0 through 1),choice(one option and a probability for each) andscore(a value on an ordered scale and a probability per level), asked about astatethat "can be text or JSON". -
Where it runs. In a Microsoft Foundry project, as a
DataZoneStandarddeployment in selected regions or aGlobalStandardone, and through OpenRouter. Microsoft's Foundry post says it is in public preview. - Price. "Input tokens cost $0.042 USD per million tokens. Output tokens are free."
- Limits Microsoft lists. Scores "can change based on how you phrase or order questions and options"; "Calibration is strongest on familiar task types"; for consequential decisions about people, "Don't use it as the sole decision-maker."
Microsoft also reports its own benchmark results, for example "the highest accuracy in our 36-benchmark comparison", measured against models it chose. None of the four Microsoft pages we read offers the weights for download: the way in is a deployment or an API call.
The request shapes are close. Microsoft's documentation says its "request pattern is similar to
a system-one API", and jevos speaks TypeSafe Jev's wire format at POST /v1/systemone. From
Microsoft's pages and our README:
| Microsoft-Decision-1 | jevos-v4 | |
|---|---|---|
| Who runs it | Microsoft, in Foundry or via OpenRouter | you, on your machine |
| Weights | not offered on the pages we read | MIT, on GitHub and Hugging Face |
| Hardware | Microsoft's | your CPU, no GPU, about 1 GB of memory |
| Question types |
noul, choice, score
|
noul, choice, score (scores early) |
criteria on a yes/no |
used for borderline cases | checked, not read |
| Authentication | Microsoft Entra ID or api-key header |
none, or a Bearer key with JEV_API_KEY
|
model in the request |
your deployment name |
jev-latest or the served model's name |
| Languages, context | not stated on the pages we read | English, 8,192 tokens |
| Price | per input token, output free | free, on your hardware |
The criteria row is the one to watch. Microsoft suggests defining true and false there to make
borderline yes/no questions more precise; jevos accepts the field and does not read it, so it
belongs in instructions, as on
LLM policy decisions: put the rule in the question.
The model name and auth header differ too; we have not run Microsoft's sample code on jevos.
The step-by-step switch, written for TypeSafe's Jev, is on
an open-source alternative to Jev.
The text stays on the machine. jevos listens on 127.0.0.1:8017 and needs no network to
answer. Microsoft's DataZoneStandard "keeps inference processing within that data zone"; that
narrows where the text goes, but it still leaves your network.
The network floor goes away. Microsoft advises creating the Foundry resource "in a region close to your application and users" to minimize network latency, because every hosted call pays a round trip before the model starts; why a hosted LLM API cannot answer in 50 ms works through it. This is about network hops, not speed: we have not timed Microsoft-Decision-1.
The cost changes shape. A per-token price grows with volume; a local model costs the same for a thousand questions or a million, and the limit becomes CPU time.
It works offline and stays put. A local model answers in CI or on a machine with no route
out (offline AI for decisions). /health reports the SHA-256 of
each model file, so a log line shows which files made a decision.
Being straight about the limits:
- No measurement against Microsoft-Decision-1. We have not run one and know of no published one.
- A hosted decision model can be more accurate. On the six tasks in our README, TypeSafe's hosted Jev scored higher than jevos-v4 on every one, for example 0.69 against 0.50 on fraud points (the full table). That says nothing about Microsoft-Decision-1; it says a local model has to be measured on your cases, not assumed equal.
- English only, 8,192 tokens of context, and early scores: the most probable level is right on 58.5% of held-out score questions, 86% within one level.
-
You operate it. Uptime and scaling are yours; Microsoft's
GlobalStandardgives "access to global capacity", jevos has the threads of one machine.
Pick Microsoft-Decision-1 if your stack already lives in Azure, if you want capacity someone else manages, or if your texts are not in English (check first: the pages we read do not list languages).
Pick jevos if the text must not leave the machine, if a network round trip hurts the loop the decision sits in, if you need it offline, or if you want the weights in hand.
Either way, test on your own labelled cases. Microsoft says the same: "Validate the model on data that's representative of your use case." Since the question types match, one test set can be asked of both, and a model cascade can keep confident answers local and send the doubtful middle to a hosted model.
Is there an open-source alternative to Microsoft-Decision-1? jevos is one: an MIT-licensed decision model that answers yes/no, choice and score questions with probabilities on a CPU. It has not been measured against Microsoft's model.
Can I download Microsoft-Decision-1's weights? None of the Microsoft pages we read offers them; they describe a Foundry deployment or OpenRouter. jevos's weights are on GitHub and Hugging Face.
Is jevos faster or more accurate than Microsoft-Decision-1? No one we know of has measured it. A local call has no network round trip, and on our six tasks the hosted Jev was more accurate than jevos-v4.
Does Microsoft-Decision-1 use the same request format as jevos? The field names in
Microsoft's documentation (state, questions, noul, choice, score, criteria) are the
ones jevos reads. The model name, the authentication and criteria on yes/no questions differ.
See also: what is a decision model in AI?, local vs hosted LLM decisions and self-hosted AI for decisions.
- jevos facts, the six-task table and the score figures: the jev README; the accuracy numbers are our own measurements.
- Microsoft's description, price, benchmark claim and OpenRouter: Introducing Microsoft-Decision-1, Command Line blog, dated 2026-10-09, fetched 2026-10-12.
- Request fields, deployments, limitations and advice: Deploy and use Microsoft-Decision-1 in Microsoft Foundry, Microsoft Learn, fetched 2026-10-12.
- Public preview: Introducing Microsoft-Decision-1 in Microsoft Foundry, Microsoft Tech Community, fetched 2026-10-12.
- No measurement of jevos against Microsoft-Decision-1 exists, ours or any we know of.
From the notes of jev, whose repository calls jevos an open-source alternative to Jev and Microsoft-Decision-1; here it is compared on what it is.
- Ask a local LLM a yes/no question and get P(yes)
- Zero-shot text classification with yes/no questions
- LLM policy decisions: put the rule in the question
- LLM as a judge on a CPU
- What is a decision model in AI?
- AI decision making with probabilities and thresholds
- Why a small LLM says yes when the answer is no
- Small LLMs and arithmetic in yes/no questions
- Our held-out benchmark said 0.855, new questions said 0.757
- Does an LLM know when a fact is missing?
- Snake AI: a local model vs a cloud API
- jevos vs Jev vs Laya for yes/no decisions
- An open-source alternative to Jev for yes/no decisions
- jevos vs the OpenAI API for yes/no classification
- jevos vs Ollama for yes/no decisions
- jevos vs bart-large-mnli for zero-shot classification
- A yes/no LLM vs a fine-tuned BERT classifier
- jevos vs SetFit: zero-shot vs few-shot classification
- jevos vs Llama Guard for content safety checks
- jev serve vs llama.cpp server for classification
- jevos vs LM Studio: a decision server, not a chat app
- Local vs hosted LLM decisions: latency, cost, privacy
- A yes/no LLM vs a business rules engine
- LLM decisions vs keyword rules and regex
- An open-source alternative to Microsoft-Decision-1
- Decision engine vs decision model
- The fastest AI model for yes/no decisions
- What makes a local LLM fast on a CPU
- Why one forward pass beats generating an answer
- Prefill vs decode: where LLM latency comes from
- Why LLM latency grows with the length of the text
- Why a hosted LLM API cannot answer in 50 ms
- Many questions about one text: why the extra ones are cheap
- CPU or GPU for a small LLM
- Latency budgets: where a 200 ms model fits
- Measuring LLM latency: median, p90 and warm-up
- Q4_K_M vs Q8_0: speed and size for a small model
- Throughput vs latency for a decision server
- OpenVINO for LLM inference on a CPU
- INT8 weight quantization explained
- What P(yes) means, and what it does not
- LLM calibration explained with yes/no answers
- Expected calibration error (ECE), explained
- Temperature scaling for LLM probabilities
- Platt scaling for a yes/no model
- Reading a reliability diagram
- How to choose a threshold for P(yes)
- Thresholds when a wrong yes costs more than a wrong no
- Human in the loop AI with a review band
- Precision and recall at a P(yes) threshold
- Base rates: why a 0.9 yes can still be wrong often
- Combining yes/no answers with AND, OR and NOT
- Logits, log-odds and P(yes)
- LLM confidence scores: probabilities vs self-reports
- How to write yes/no questions an LLM answers well
- Negation in yes/no questions for an LLM
- One condition per question: splitting compound questions
- Ask whether the text says it at all
- Scores as yes/no thresholds: is it at least high?
- Sending JSON as the text: designing the state
- Why wording changes an LLM's answer, and how to test it
- Mainly about: questions for messages with several topics
- Yes/no questions about tone and emotion
- Asking about intent: what does the writer want?
- Yes/no questions about long documents
- Using an English-only LLM with other languages
- Multi-class text classification with a choice question
- Content moderation with a local LLM
- A Discord moderation bot with a local LLM
- Spam detection with yes/no questions
- Review moderation with a local LLM
- Email triage with a local LLM
- Support ticket routing with yes/no questions
- Urgency detection in customer messages
- Sentiment analysis with yes/no questions
- Intent detection with a local LLM
- Lead qualification with yes/no questions
- Fraud case triage with a local LLM
- Phishing email screening with a local LLM
- Log and alert triage with a local LLM
- Checking text for personal data with yes/no questions
- Prompt injection screening with a small model
- Document classification with a local LLM
- Product categorization with yes/no questions
- Contract clause detection with a local LLM
- Refund request triage with a local LLM
- Detecting cancellation intent in customer messages
- Reranking search results with yes/no questions
- RAG evaluation with yes/no questions
- RAG faithfulness check with a local LLM
- Hallucination detection with a local LLM
- LLM regression tests in CI with yes/no checks
- Rubric design for an LLM judge
- Pairwise comparison with a yes/no judge
- LLM judge bias and how to control it
- Evaluation metrics for yes/no classifiers
- Building a yes/no test set for your own data
- Accuracy by kind of question: why one number hides failures
- Generating test questions with answers computed by code
- Benchmark contamination and truly held-out tests
- An LLM router with yes/no questions
- A model cascade: small model first, large model on doubt
- Semantic routing vs yes/no questions
- Gating AI agent tool calls with yes/no checks
- AI agent guardrails with yes/no questions
- Stop conditions for AI agents
- Logging LLM decisions for audit
- Reducing LLM cost with local yes/no decisions
- Replacing chat LLM calls with yes/no questions
- Structured output vs a probability
- A Python client for local LLM decisions
- Calling a local LLM decision server from JavaScript
- Local LLM yes/no decisions in n8n
- A Slack bot that uses local LLM decisions
- Home Assistant automations with local LLM decisions
- A LangChain tool for local yes/no decisions
- Batch decisions from files with jev decide
- Running LLM yes/no checks in GitHub Actions
- Securing a local LLM server with an API key
- curl examples for a local LLM decision API
- Self-hosted AI for decisions
- A private LLM for text classification
- On-premise LLM for business decisions
- GDPR and automated decision-making with an LLM
- Offline AI for decisions: no network needed
- Edge AI decisions on a CPU
- Run an LLM locally without a GPU
- Small language models explained
- When a small model is enough, and when it is not
- An LLM on a laptop: what it can do in real time
- What is GGUF, for someone deploying a classifier
- GGUF quantization types explained: Q4_K_M, Q8_0 and others
- GGUF vs safetensors
- llama.cpp vs Ollama for a classification service
- llama-cpp-python vs calling llama.cpp through ctypes
- llama.cpp on Windows without compiling
- Running llama.cpp CPU only
- Using llama.cpp prebuilt binaries instead of building
- Download a GGUF model from Hugging Face