Industry News
Microsoft Decision-1.1 is specifically designed for corporate decision-making purposes. If you’re unsure about a particular option, you can simply opt to abstain from making a choice. There are no fees associated with using this tool.
The model released by Microsoft this time does not chat with people, but only takes multiple-choice questions.
On November 17, Microsoft launched the decision-making model Microsoft-Decision-1.1, which has been launched on Microsoft Foundry, OpenRouter and Vercel. It targets those high-frequency and trivial judgments in enterprise software: where should requests be routed, who should assign work orders, whether content should be released, and where the workflow should go next.
The way it works is restrained. The developer first defines the rules and allowed operations, and the model only selects answers within this range and gives a calibrated confidence score. When the information is insufficient, it can choose to give up or prompt for manual processing.
For enterprises, this is more important than running scores.
In critical business processes, a model that confidently answers the question incorrectly is much more dangerous than a model that admits to being unsure. Decision-1.1 makes "uncertain" a formal output, which is equivalent to leaving a hidden interface for people.
In terms of performance, it scored 68 points on the Artificial Analysis Decision Index v2.1 benchmark, which is 6 points higher than the previous generation and ahead of opponents such as Jev 1.14, Google Gemini Decision 1 and OpenAI GPT-6 Luna. The speed is more than twice as fast as the previous generation. The price is very low, USD 0.042 per million input tokens, and output is free. Microsoft also customized its inference engine and optimized it based on the medium-sized hybrid expert model.
Since its public preview, more than 4,000 customers have used it. The scenarios include model routing, customer service ticket offloading and business process automation. Early users include Epic and Raiffeisen Bank. The model also supports multi-modal input such as photos and screenshots.
Microsoft itself also uses it. It is used internally for Copilot for request routing. The accuracy is equivalent to that of a large language model, but the speed is 2.2 times faster and the cost is 4.4 times lower.