Jev vs Microsoft-Decision-1: same job, bigger name

Both models answer the same kind of question: you send an input and typed questions, and get back an answer with a probability for each option, with no text written. If you can run a job on one, you can usually run it on the other with small changes to the request.

The differences that are documented are about where and how you run it. Jev is TypeSafe's service, called through its API or a gateway. Microsoft's model is deployed in Microsoft Foundry on Azure. The one independent test we found, Artificial Analysis's Decision Index, scores Jev higher (57.1 for Jev 1.14 and 57.5 for Jev 1.13, against 53.8), but it is one index of four tests, so check accuracy on your own data.

From Microsoft's documentation page, updated on 9 October 2026:

• The model exists in Microsoft Foundry's catalogue, as version 1, and is meant for classification, routing, ranking, grading and binary decisions.
• It uses three question types, named noul (yes/no), choice and score, and returns a probability for each option rather than text.
• The request goes to an address ending in /systemone, and the documentation says the pattern is similar to a system-one API: you supply a state and bounded questions.
• You deploy it with an Azure subscription, in a global deployment or in a data zone that keeps processing within a region.
• It lists limitations, including that scores can change with how you phrase or order the questions, and that calibration is strongest on familiar task types.

MarkTechPost (9 October) and Help Net Security (12 October) report that the model is post-trained from Alibaba's open-weight Qwen3.5-9B. Satya Nadella announced it on X on 9 October.

Sources disagree on whether it is generally available or a public preview. Microsoft's own documentation page still does not state a price, but OpenRouter's own listing for the model now confirms $0.042 per million input tokens with free output, the same list price as Jev.

Everything in this section is Microsoft's own claim, as Help Net Security reports it, and none of it has been independently checked. Microsoft says it tested on 36 benchmarks covering nearly 150,000 questions kept out of training, and that the model is about 35 times faster than GPT-6 Sol on median latency and 2.5 times faster than H2O-Lightning-4B v1.1, the next fastest. It also describes internal tests by its Xbox, Copilot and Discovery teams.

Some coverage says Microsoft's launch material compared the model with Jev 1.13 and reported slightly higher average accuracy for its own. We could not read that table, so we do not quote its numbers here. Artificial Analysis's independent index, below, does not show that lead.

Artificial Analysis, an independent evaluator, launched a Decision Index on 13 October that scores decision models on four tests it runs itself. On its leaderboard (read 18 October), Microsoft-Decision-1 scored 53.8, against 57.5 for Jev 1.13 and 57.1 for Jev 1.14. Liquid AI's d1-ultra led at 65.2. The leaderboard lists Microsoft-Decision-1 under OpenRouter, which fits press reports that it is available there.

That is one independent index, days old, with four tests that are not your task, so it does not settle which model is better for you. But it does not support the accuracy lead that Microsoft's own table claims over Jev.

Based on what is documented today:

• Choose Jev if you want TypeSafe's API from any cloud or through a gateway, you want the model that has been available since 15 September, or you already use it.
• Look at Microsoft-Decision-1 if you already build on Azure and want the model deployed, billed and secured there, or you need to choose a region or data zone.
• Test both before you commit. The request shapes are close, so you can send the same labelled sample of your own data to each and compare accuracy, speed and cost.

More on how Jev compares, and where it fits:

• Jev vs OpenAI's Decisions API: who wins?
• Jev vs open models: can you skip Jev?
• Jev cost calculator