Back
FR

Model creator

Microsoft

Every Microsoft model in the BenchSift catalogue, with its scores, ranks, prices and capabilities as the sources publish them.

Hugging Face · microsoft
View in the catalogue
Models
19
Open weights
6
6 of 19 models
Latest release
Oct 9, 2026
Microsoft-Decision-1 · First dated model: Apr 16, 2024
Median price
$0.42
$0.0315 to $3.75 per million tokens, across 8 priced models

Best scores and ranks

The creator's best model on each measure, and its rank among every measured model in the catalogue.

Intelligence

Phi-4 Mini Instruct
6.3Rank 644 of 781 measured models

Coding

Phi-4 Mini Instruct
3.8Rank 292 of 300 measured models

Math

Phi-4
18.0Rank 215 of 270 measured models

Output speed

Phi-4
34.6 t/sRank 429 of 449 measured models

Lowest price

Microsoft-Decision-1
$0.0315Rank 41 of 740 measured models

OpenRouter ranking

Microsoft-Decision-1
#124

Capabilities

Number of models that declare each capability. Tool calling and structured outputs come from OpenRouter.
Reasoning2 of 19
Open weights6 of 19
Tool calling0 of 19
Structured outputs2 of 19

Input

  • Text17
  • Image5
  • Audio3

Output

  • Speech8
  • Text7
  • Image3
  • Transcription2
  • Decisions1

Overview

Model types

  • Audio8
  • Text7
  • Image3
  • Decisions1

Releases by year

  • 202610
  • 20252
  • 20243

Hugging Face

Official repositories confirmed by Hugging Face metadata.

3 official repositories

Licenses

  • mit3

Models

19

Reasoning levels are folded into one entry. The rank is the whole catalogue's: sorting does not change it.

Where these figures come from

Each catalogue model is tied to its creator by the identity Artificial Analysis publishes, or OpenRouter's and Models.dev's for models it does not track. Counts fold the reasoning levels of one model together. Ranks compare each measure with every measured model in the catalogue, configurations included, as model pages do. The price is the published blended price (3 input tokens to 1 output token).

Figures computed by BenchSift from the catalogue: Artificial Analysis, OpenRouter, Models.dev and Hugging Face. A value no model carries is not shown.

Data retrieved on Nov 12, 2026.

Preferences and analytics

Rybbit: visits, performance, clicks, copies, forms and session replay (masked inputs), with your permission. Refusing does not limit the site.

Privacy

Allow navigation analytics?

With your permission, Rybbit measures visits, performance and interactions (clicks, copies, forms), including session replay, to improve BenchSift. Input values are masked in replays. Refusing does not limit the site.

Your choice is stored for 180 days. Change it at any time using “Privacy preferences” in the footer. Withdrawal reloads the page.

No choice stored: Rybbit is disabled.

Privacy