WED, NOVEMBER 18, 2026
Independent · In‑Depth · Practitioner‑Tested
Large Language Models

Microsoft-Decision-1.1 vs Gemma 4 (2026): The Hosted Router and the Open Model Under It

Microsoft-Decision-1.1 is built on Gemma 4 26B A4B and scores 68 on the Artificial Analysis Decision Index v2.1. Gemma 4 is open-weight and self-hostable. Which you want depends on whether image inputs and a managed endpoint are worth giving up control of the weights.

🕐 6 min read 👁 74 views 📅 Nov 18, 2026
The short version
● Same lineage Microsoft's Book of News states Decision-1.1 is built on Gemma 4 26B A4B.
● Decision-1.1 adds Image inputs, a published score of 68 on the Artificial Analysis Decision Index v2.1, and managed availability on Foundry, OpenRouter and Vercel.
● Gemma 4 gives you The weights. Self-hosting, fine-tuning, and no data leaving your infrastructure.
● Not comparable yet Cost. Microsoft published no price for Decision-1.1.

Routing models are the least glamorous part of an AI stack and increasingly the part that decides what it costs. Every request has to be sorted before it is answered, and doing that with a frontier model means paying frontier prices to discover that a question was trivial.

Microsoft-Decision-1.1 is Microsoft's answer, and it is unusual in that Microsoft has said what it is built on: Gemma 4 26B A4B, from Google's open-weight family. That turns a vague build-or-buy question into a concrete one, because the base model is downloadable.

What the hosted version adds

Three things, on Microsoft's own account. Image inputs, which version 1 did not have - routing decisions can take a screenshot or a scanned document into account rather than text alone. A score of 68 on the Artificial Analysis Decision Index v2.1, from an independent evaluator. And distribution: generally available in Microsoft Foundry, OpenRouter and Vercel, so it drops into most existing routing setups without new infrastructure.

There is a fourth thing that is harder to quantify and probably matters more. Microsoft uses it for Copilot routing and for Cowork's Auto mode. Whatever its benchmark says, it is handling production traffic at Copilot volume, which is a stronger signal than a score.

What the base model gives you

Gemma 4 26B A4B is open-weight. 26 billion parameters with roughly 4 billion active per token is small enough to serve cheaply on your own hardware and fast enough to sit in front of every request, which is the whole point of a routing model.

Self-hosting buys three things the hosted version cannot. The routing decision never leaves your infrastructure, which matters when the thing being classified is the sensitive part. You can fine-tune on your own taxonomy rather than a general one - routing is exactly the task where a domain-specific model beats a general one. And at high volume you pay for compute rather than per request.

Where the comparison breaks down

Cost. Microsoft published no per-token price for Decision-1.1 at Ignite, and no price for any model in that announcement. Until it does, the one number that would decide most of these choices does not exist. Anyone presenting a cost comparison between these two right now is estimating.

How to choose

Need image inputs in routing, or want it working this afternoon: the hosted model, on whichever of Foundry, OpenRouter or Vercel you already use. Routing on data that cannot leave your estate, or routing on a taxonomy that is yours rather than generic: the open weights. Very high request volumes where per-request pricing compounds: self-host, and revisit if Microsoft publishes a price that undercuts your own compute.

FAQ

Is Microsoft-Decision-1.1 the same as Gemma 4?
No, but it is built on it. Microsoft's Ignite 2026 Book of News states Decision-1.1 is built on Gemma 4 26B A4B, and adds image inputs over its own version 1.
What does Microsoft-Decision-1.1 score?
68 on the Artificial Analysis Decision Index v2.1, per Microsoft's announcement.
Can I self-host Gemma 4 26B A4B?
Yes, it is part of Google's open-weight Gemma family. Check the current licence terms before commercial deployment.
Which is cheaper?
Not answerable yet. Microsoft has published no price for Microsoft-Decision-1.1.
Where can I use Microsoft-Decision-1.1?
It is generally available in Microsoft Foundry, OpenRouter and Vercel.
⚖ Our Verdict

Microsoft-Decision-1.1 if you want a routing endpoint today and image inputs matter - generally available on Foundry, OpenRouter and Vercel, scoring 68, and already routing Copilot at scale. Gemma 4 26B A4B if the data cannot leave your estate, you need your own taxonomy, or volume makes per-request pricing hurt. No winner: Microsoft published no price.