Skip to content
AI.info

The Pulse

Microsoft makes Decision-1.1 generally available with image support

Microsoft-Decision-1.1 is generally available in Microsoft Foundry and through OpenRouter and Vercel. Microsoft says the model is more than twice as fast as its predecessor at the same cost and can accept images in requests.

Microsoft makes Decision-1.1 generally available with image support

Microsoft-Decision-1.1 is now generally available in Microsoft Foundry and through OpenRouter and Vercel, giving developers access to an updated model for structured decisions in software. Microsoft says more than 4,000 customers have used it since its public preview began last month.

The updated model can accept images, including photos and screenshots, as part of a request. It is designed for routing, classification, prioritization, verification and workflow control, and is available for enterprise production deployment.

Decisions within set limits

Microsoft designs the decision models to choose from permitted actions under constraints set by developers. They provide confidence scores that applications can use to defer a decision when uncertainty is high, according to the company.

The model is intended to make structured decisions in software rather than handle every task as a general-purpose language model would. Its training includes distinguishing closely related choices, recognizing uncertainty and deciding when to answer or abstain, Microsoft says.

The company also says it trained the model to give consistent answers in similar scenarios, report confidence, divide confidence when more than one answer is valid, and abstain when information is incomplete. Developers can set thresholds for acting, flagging or escalating a decision.

Training and infrastructure

Microsoft based the model on Gemma 4 26B A4B, a mid-sized mixture-of-experts model. The company lists A100, H100/200 and GB200/300 among the hardware on which it can run.

The company trained the model using supervised fine-tuning and reinforcement learning across hundreds of synthetic and licensed datasets. Those datasets span tens of thousands of types of tasks across a range of industries, according to Microsoft.

A custom inference engine is optimized for the computational demands of decision models, rather than the token-by-token text generation typical of traditional large language model inference engines. Microsoft says the engine routes requests across its fleet of GPUs to reduce latency and cost.

Speed and benchmark claims

Microsoft says Decision-1.1 is more than twice as fast as its first decision model at the same cost, alongside improvements in intelligence and calibration. The company also says its decision models are significantly faster and less expensive than general-purpose large language models on decision-making tasks.

On the Artificial Analysis Decision Index v2.1, the new model scored 68, six points higher than Microsoft's first model, the company says. The index measures model quality, calibration and latency.

Microsoft says the score also puts the model ahead of Jev 1.14, Google's Gemini Decision 1 and OpenAI's GPT-6 Luna (Decisions) on that index. The benchmark comparison is distinct from the company's speed comparison with its earlier model.

Source

  1. Microsoft

Companies in this story

Related stories

Explore

More articles