Microsoft is already experimenting with its newly announced Microsoft-Decision-1 model in both Bing Search and Copilot, less than a week after unveiling the decision-scoring system. Bing Corporate Vice President Krishna Madhavan confirmed the tests when asked directly, replying on X: "we are always experimenting!"

The confirmation separates fact from speculation in two distinct ways. Madhavan's statement — from a named Bing executive — confirms experimentation exists. What remains unconfirmed is the scope: there is no evidence Bing applies the model to all queries or all users, and no tool-signal data yet shows measurable ranking shifts tied to it.

What is Microsoft-Decision-1?

Microsoft announced the model last week and made it available in Microsoft Foundry and through OpenRouter. The company described it as "a decision-scoring model delivers top performance in latency and quality on structured decision tasks to outperform both LLMs and other decision models."

Unlike a general-purpose language model, Microsoft-Decision-1 targets specific operational jobs inside larger systems:

• routing
• classification
• prioritization
• verification
• workflow control

Microsoft's positioning statement emphasizes integration: the model is "designed for routing, classification, prioritization, verification, and workflow control, making it easier to incorporate decision intelligence into existing applications, agents, and workflows in a secure, trusted environment."

In practical terms, a model like this would not replace the LLM generating search answers or Copilot responses. It would sit in the plumbing — deciding which engine handles a query, how to classify intent, which results to verify or prioritize, and how to orchestrate multi-step agent workflows.

Who confirmed what?

Rangan Majumder, a Microsoft leader working on Copilot, posted on X: "We're experimenting with it in Copilot and it's pretty good! Very impressed with how quickly they improved it as well."

That is an on-record statement from a Microsoft executive confirming Copilot testing, with a qualitative assessment attached — though no benchmark numbers or user coverage figures accompanied it.

Madhavan, who leads Bing, offered a more general confirmation when asked whether Bing Search itself uses the model. His "we are always experimenting" reply neither confirmed nor denied deployment at scale. That gap matters for anyone tracking whether Microsoft-Decision-1 currently influences live Bing rankings.

So the confirmed record stands as follows:

• Microsoft announced Microsoft-Decision-1 last week and shipped it via Microsoft Foundry and OpenRouter.
• Majumder confirmed Copilot experimentation with the model.
• Madhavan confirmed Bing experiments in general terms, without specifying the model's role in ranking.
• No query coverage data, latency benchmarks, or ranking-impact measurements have been published.

Why a decision-scoring model matters for search

Microsoft's claim that the model can "outperform both LLMs and other decision models" on structured decision tasks points to a specific engineering tradeoff: speed versus capability. Routing and classification decisions inside a search or agent pipeline happen millions of times per hour, and using a full LLM for each one adds latency and cost.

A dedicated decision-scoring model that beats LLMs on those tasks at lower latency could change how query routing works — for example, deciding faster whether a query should hit the index, an AI-generated answer, or an agentic workflow. Microsoft claims top performance on both latency and quality for structured tasks, though the company has not published comparative benchmark tables alongside the announcement.

For site owners and search professionals, the near-term impact is indirect. If Microsoft-Decision-1 takes over routing and prioritization inside Bing and Copilot, it could eventually affect which content sources feed AI-generated answers and how quickly verification steps run — but no data yet connects the model to observable changes in Bing results or Copilot citations.

What to watch next

The signals worth monitoring are concrete: Bing's official blog or release notes for any deployment announcement beyond "experimenting," benchmark publications comparing Microsoft-Decision-1 against LLM baselines on routing and classification, and any statements at upcoming Microsoft events about where the model sits in the Bing and Copilot stack. Madhavan's noncommittal phrasing suggests full rollout remains unannounced, so tool-observed ranking fluctuations should not be attributed to this model without corroborating evidence.