It was introduced by Achint Srivastava, VP of Software Engineering in the Office of the CTO at Microsoft, on the Command Line blog in October 2026.

How much does it cost?

No pricing information is provided on the announcement page; check Microsoft Foundry or OpenRouter for current pricing.

Microsoft-Decision-1 — A fast AI model for structured decision-making

Microsoft-Decision-1 is a decision-scoring model from Microsoft, announced on the Command Line blog, designed to deliver structured outputs that software can act on immediately. Unlike LLMs built for text generation and complex reasoning, it is purpose-built for fast decision-making — routing, classification, prioritization, verification, and workflow control — and is available in Microsoft Foundry and through OpenRouter. Microsoft claims it achieved the highest accuracy in its own 36-benchmark comparison (spanning nearly 150,000 questions from benchmarks kept blind from training) and was the fastest model measured, 2.5x quicker than runner-up H2O-Lightning-4B v1.1 and 35x quicker than GPT-6 Sol.

Key Features

Structured decision outputs

Purpose-built to produce structured outputs that software can immediately act on, rather than generating free-form text like an LLM.

Decision intelligence for applications and agents

Designed for routing, classification, prioritization, verification, and workflow control, so decision logic can be incorporated into existing applications, agents, and workflows.

Low-latency performance

In Microsoft's benchmarking it was the fastest model measured: 2.5x quicker than H2O-Lightning-4B v1.1 and 35x quicker than GPT-6 Sol.

Top accuracy on structured decision tasks

Achieved the highest accuracy in Microsoft's 36-benchmark comparison covering nearly 150,000 questions across benchmarks kept blind from training.

Available in Microsoft Foundry and OpenRouter

Hosted in Microsoft Foundry — Microsoft's secure, trusted enterprise AI environment — and also accessible through OpenRouter.

How It Works

Score or classify an input

Send a task to the model (e.g., a routing, classification, prioritization, or verification request) and it returns a structured decision output suitable for programmatic use, rather than free-form generated text.

Integrate into your workflow

Wire the model's outputs into existing applications, agents, or workflow-control logic, using it as the low-cost, low-latency decision layer behind larger systems.

Pros & Cons

Pros

Specialized for fast, structured decision tasks rather than text generation, which suits routing, classification, and verification workloads
Highest accuracy in Microsoft's 36-benchmark comparison (~150,000 questions on blind benchmarks)
Fastest model measured in Microsoft's benchmarks: 2.5x quicker than H2O-Lightning-4B v1.1 and 35x quicker than GPT-6 Sol
Available in two places — Microsoft Foundry for enterprise integration and OpenRouter for broader access
Runs within Microsoft's secure, trusted environment for enterprise deployments

Cons

Who It's For

Best for

Engineering teams building applications, agents, or workflows that need fast routing, classification, prioritization, or verification decisi
Teams wanting a low-cost, low-latency decision layer that complements — rather than replaces — LLMs
Organizations already using Microsoft Foundry that want decision intelligence in a secure, trusted environment
Developers using OpenRouter who want a specialized decision model among their available options

Not ideal for

Teams needing open-ended text generation or long-form reasoning — LLMs remain the right tool for that

FAQ

What is Microsoft-Decision-1?

It's Microsoft's decision-scoring model, purpose-built to produce structured outputs that software can act on immediately, designed for routing, classification, prioritization, verification, and workflow control.

How is it different from an LLM?

Unlike LLMs, which generate text or reason through complex problems, decision models are built to deliver structured outputs for machine consumption, at very low cost with high performance on decision and classification tasks.

How fast and accurate is it?

In Microsoft's benchmarking it achieved the highest accuracy in a 36-benchmark comparison covering nearly 150,000 questions from benchmarks kept blind from training, and it was the fastest measured: 2.5x quicker than H2O-Lightning-4B v1.1 and 35x quicker than GPT-6 Sol.

Where is it available?

Microsoft-Decision-1