Live24 models · 21 open source · avg 36.1
Microsoft logo

Microsoft

🇺🇸United StatesWebsite
Top Model
phi-3-small 7.4B
avg score
24
Total Models
tracked on BenchGecko
21
Open Source
88% of models
$0.07
Cheapest Model
per 1M input tokens
36.1
Avg Benchmark
across 11 scored models

Model Categories

LLM24

Pricing Range $/1M input tokens

$0.07
$0.62
Low: $0.07Median: $0.62High: $0.62

Open Source Ratio

88%
21 open source3 proprietary
#ModelAvgaider editaider poly?ANLI?APEX-Agents?ARC AI2?ARC-AGI?ARC-AGI-2?aa agentic?aa coding ?aa qualityseal audioseal audioseal audioBalrog?BBH?hf bbhC-EvalCadEval?charxiv re?charxiv re?arena elo arena elo chess puzz?CMMLUCSQA2Cybench?deepresear?EnigmaEvalfiction li?Fortressfrontierma?frontierma?GeoBench?GPQAGPQA diamond?graphwalks?GSM8K?GSO-Bench?HellaSwag?HELM · GPQAhelm ifevahelm mmlu helm omni helm wildbHLE?hle toolsseal humanseal humanIFEvaljp jcommonJHumanEvalJMMLUJNLIJSQuADLAMBADA?lech mazur?livebench livebench livebench livebench livebench livebench livebench livebench jp overallMASKMATH level 5?MATH Level 5MCP AtlasMMLU?MMLU-PROMMMLUmmmlu armmmlu bnmmmlu zhmmmlu frmmmlu demmmlu himmmlu idmmmlu itmmmlu jammmlu kommmlu ptmmmlu esmmmlu swmmmlu yoseal multiMultiNRCMUSROpenBookQA?oc aime202oc gpqa dioc hleoc ifevaloc livecodoc mmlu prOSWorld?otis mock ?PIQA?posttrainbseal pro rseal pro rseal propeseal remotScienceQA?SciPredictSimpleBench?simpleqa v?seal swe aseal swe aswe bench swe bench swe bench seal swe bseal swe bswe bench ?swe bench ?terminal b?the agent ?TriviaQA?TutorBenchUSAMOVideoMME?VISTAseal visuaVPCT?WeirdML?Winogrande?$/1M inContextReleased
1Microsoft logophi-3-small 7.4B🇺🇸 MicrosoftOpen78.8--37.1-87.6---------72.1-----------------------69.3------------------------------67.6-------------------84.0-----------------------------58.1-------63.0N/A-Jan 242y ago
2Microsoft logophi-3-medium 14B🇺🇸 MicrosoftOpen69.0--33.7-88.8---------75.2-------------------3.5---76.5---------------------------17.6--70.7-------------------83.2-----------------------------73.9-------63.0N/A-Jan 242y ago
3Microsoft logophi-3-mini 3.8B🇺🇸 MicrosoftOpen68.3--29.2-79.9---------62.3-----------------------68.9------------------------------58.4-------------------84.0-----------------------------64.0-------41.6N/A-Jan 242y ago
4Microsoft logoWizardLM-2 8x22B🇺🇸 MicrosoftOpen61.744.4--------------48.6-----------------17.6--------------52.7------------------25.0--40.0-----------------14.5---------------------------------------$0.6266KApr 242y ago
5Microsoft logoPhi 4🇺🇸 MicrosoftOpen54.2-------0.011.210.4---11.6-55.3-----1255.4-----------11.541.4-------------68.8------62.6----------64.950.0-79.748.6-----------------10.1--------13.7------------------------------$0.0716KJan 251y ago
6Microsoft logoPhi 3.5 Mini Instruct🇺🇸 MicrosoftOpen51.5---------------36.8-----------------12.0--------------57.8------------------19.6--32.9-----------------10.1---------------------------------------N/A-Aug 241y ago
7Microsoft logoPhi 3 Mini 4k Instruct🇺🇸 MicrosoftOpen51.1---------------36.6-----1127.2-----------11.0--------------54.8------------------16.4--33.6-----------------13.1---------------------------------------N/A-Apr 242y ago
8Microsoft logoPhi 4 Mini Instruct🇺🇸 MicrosoftOpen48.9-------2.73.68.4-----38.7-----------------7.9--------------73.8------------------17.0--32.6-----------------6.5---------------------------------------N/A-Feb 251y ago
9Microsoft logoPhi 2🇺🇸 MicrosoftOpen31.0--13.8-67.9---------45.928.0-----------------2.9----38.1---------27.4------------------3.0-44.518.1-----------------13.864.8-----------------------------45.2-------9.4N/A-Dec 232y ago
10Microsoft logoPhi-1.5🇺🇸 MicrosoftOpen15.6----25.9----------7.5-----------------2.4----30.1---------20.3------------------1.8-16.87.7-----------------3.416.3-------------------------------------46.8N/A-Jan 242y ago
90+ Gold 80-89 70-79 60-69 <60Scores in % unless noted. Avg = unweighted mean across tested benchmarks.

Quick answers · sourced from our data

How many models does Microsoft have?

BenchGecko tracks 24 models from Microsoft, of which 21 (88%) are open source. Every entry is updated daily from live provider feeds.

What is the best model from Microsoft?

phi-3-small 7.4B is currently the highest scoring Microsoft model we track, with an average benchmark score of 67.4. Scores are computed across every public benchmark we have data for.

What is the cheapest Microsoft model?

The cheapest Microsoft model on BenchGecko starts at $0.07 per 1M input tokens. Pricing is pulled from OpenRouter and cross-checked against official provider rate cards.

How does Microsoft compare on benchmarks?

Microsoft models average 36.1 across the benchmarks we track · see the All Providers page for the full ranking by model count, open source ratio, and average score.

Where is Microsoft based?

Microsoft is headquartered in United States. BenchGecko groups providers by region to make it easy to compare US, EU, China, and Rest of World markets.

Is Microsoft open source?

21 of 24 Microsoft models are open source (88%). The rest are proprietary · closed weights served via API.