Microsoft-Decision-1 scores fixed choices, and Microsoft says it is 35 times quicker than GPT-6 Sol

Important

Lab claim

On October 9, Microsoft released a model that returns a score for each allowed answer instead of writing text. On Microsoft's tests it was the fastest of the set, 35 times quicker than GPT-6 Sol.

How to read this claim

Measurement — The announcement does not give a parameter count. The speed and cost figures are Microsoft's.
Bill — List input and output per million tokens on the vendor page — a loop pays both again.
Door — Status on this card: Shipped.

Keep the habit · field card with prices and worked checks

What changed

A general model writes the answer. Microsoft-Decision-1 scores a list the software already wrote, in one pass. Microsoft says that job was 35 times quicker than GPT-6 Sol on its tests.

What it introduces

A model that returns a score for each allowed answer, instead of writing a paragraph. It is the wrong tool when the product needs prose.

Who benefits Teams whose agent spends most of its calls classifying, routing, or checking a step against a rubric. Anyone comparing a general model with a specialist on a repeated decision.

Try this

Take one high-volume choice your product already scores, run Microsoft-Decision-1 beside the model you use now, and keep both the accuracy and the wait. Treat 35 times as Microsoft's number until that test is yours.

What we do not know

The announcement does not give a parameter count. The speed and cost figures are Microsoft's.

Evidence

Primary Microsoft. Reported by Microsoft. The speed and accuracy comparisons are Microsoft's.

Independent No independent rerun is cited. The speed comparison and the Xbox study are Microsoft's. The announcement page does not give a parameter count.

Sources

Microsoft Primary · October 9, 2026

Still unclear? · Start here · ranks, prices, unknown The announcement does not give a parameter count. The speed and cost figures are Microsoft's.

The record

What happened

On October 9, 2026, Microsoft introduced Microsoft-Decision-1, a model for routing, classification, prioritization, verification, and workflow control. It is available in Microsoft Foundry and through OpenRouter. Instead of writing an answer, it takes a fixed set of options and returns a score for each one in a single pass. Microsoft says that in its own comparison the model had the highest accuracy across 36 benchmarks and nearly 150,000 questions held out from training. Microsoft also says it was 4.5 times quicker than Quyet-1.0-Large, the next fastest model in that test, and 35 times quicker than GPT-6 Sol. The median latency comparison with GPT-6 Sol is Microsoft's as well.

Why it matters

An agent loop often asks a general model to pick from a list: which queue, which label, whether a step is allowed. That call is where time and money go. A model built only to score those choices can sit in front of the expensive model and handle the repetitive branch. Microsoft's 35-times figure is the reason to measure it, and it is Microsoft's figure, on Microsoft's tests. It does not say the model can replace GPT-6 Sol at writing, coding, or a long investigation. Coinbase's fraud model and this one are the same kind of bet: take one scored decision off the generalist. The percentages do not transfer from one job to the other.

Microsoft describes an internal Xbox study that sorted more than 10,000 pieces of feedback into themes researchers had already defined. Microsoft says the model was competitive on quality with GPT-6 Sol, more than 14 times faster, and 200 times less expensive on that job. That study is Microsoft's.

The post says the model is for decisions software can act on immediately. It does not publish a parameter count on the announcement page.