Key points
- Microsoft Decision-1 returns calibrated probabilities for fixed answer options rather than generating free-form text.
- The model is priced at 4.2 cents per million input tokens with no charge for output tokens.
- Microsoft claims Decision-1 is 35 times faster than GPT-6 Sol on specific benchmarks.
Microsoft has launched Decision-1, a new artificial intelligence model designed to return probabilities for predefined answers instead of generating free-form text. According to InfoWorld, this approach allows software applications to use outputs directly for tasks like classifying requests or prioritizing incidents. Microsoft states that Decision-1 handles these decisions with lower costs and latency than general-purpose large language models.
How it unfolded
- Microsoft announced Decision-1 is available through Microsoft Foundry and OpenRouter, built by post-training the Qwen3.5-9B model.
- The company reported that Decision-1 achieved the highest accuracy across 36 benchmarks covering nearly 150,000 blind questions.
- Microsoft claimed the model is 35 times faster than GPT-6 Sol and 2.5 times faster than the runner-up in speed tests.
Who is affected
System administrators and developers can use Decision-1 to classify inputs, rank work, or verify AI-generated responses. According to Computerworld, the model supports yes-or-no judgments, choices among predefined options, and rubric-based assessments of agent actions. Unlike general LLMs that produce natural-language explanations, Decision-1 returns probabilities for available answers in a single pass. This design allows developers to determine the next action in an AI agent workflow without processing verbose outputs.
Microsoft lists the price at 4.2 cents per million input tokens, with no charge for output tokens. In an incident-prioritization example, a request using 86 input tokens and two output tokens cost approximately $0.0000036. At that rate, one dollar would cover more than 275,000 similar decisions. These benchmark results and cost calculations are Microsoft’s claims and have not been independently verified by third parties.
Charlie Dai, vice president and principal analyst at Forrester, noted that specialized decision models could reduce the cost and latency of agentic AI workflows. Microsoft said it plans to rebase Decision-1 on other models, including Microsoft AI (MAI) and OpenAI models, in the future. The current version relies on the post-trained Qwen3.5-9B foundation.
What happens next
Microsoft said its Decision-1 model can evaluate multiple decisions in parallel within a single request. This capability allows applications to handle complex routing or classification tasks more efficiently. Developers can integrate the model via Microsoft Foundry or OpenRouter to replace general LLM calls for specific decision-making steps.
The company intends to expand the model’s foundation beyond Qwen3.5-9B. Microsoft stated it plans to rebase Decision-1 on other architectures, including Microsoft AI and OpenAI models. This strategy suggests a focus on optimizing performance across different underlying technologies while maintaining the fixed-set probability output format.
Background: Large language models
Large language models act as probabilistic optimizers that predict the next most likely token rather than verifying logical correctness. This mechanism means they excel at pattern matching but fail at rigorous reasoning. You must treat their output as a draft requiring strict validation, not a final solution.
Read the full guide: Large Language Models as Optimizers: Why Your Code Changes Break
What to do and how to stay safe: Microsoft
- Review current AI workflows to identify tasks that use general LLMs for simple classification or routing decisions.
- Test Decision-1 on non-critical workloads to compare latency and accuracy against existing large language models.
- Monitor token usage and costs when integrating fixed-set probability models into production environments.
- Verify that output probabilities meet your application’s confidence thresholds before automating critical decisions.
Step-by-step guide: Mobile Device Security for Small Business: Control Without Cost
General security guidance from the Circuit Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is Microsoft Decision-1?
Decision-1 is an AI model that returns calibrated probabilities for predefined answers instead of generating free-form text, targeting tasks like classification and prioritization.
How much does Decision-1 cost?
Microsoft lists the price at 4.2 cents per million input tokens, with no charge for output tokens.
Is Decision-1 faster than GPT-6 Sol?
Microsoft claims Decision-1 is 35 times faster than GPT-6 Sol and 2.5 times faster than the runner-up in its benchmarks.
