Benchmark Almanac

Microsoft

Best on the Capabilities Index Phi-4 129.0
31 models 18 model cards read microsoft.com/ai

How does Microsoft stack up?

Every model on Epoch AI's Capabilities Index by release date, with Microsoft's in color.

50 100 150 2023 2024 2025 2026 2027 Phi-4 phi-3-medium 14B Phi-2 Phi-1.5 50 100 150 2023 2024 2025 2026 2027 Phi-4 phi-3-medium 14B

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Microsoft's best models

Microsoft Dec 2024

Phi-4

Open weights Small

129.0 #159 of 256
Epoch Capabilities Index
Self-Reported Index 125.1 #313 of 404
  • GPQA Diamond 56.1% #144 /229
  • BigCodeBench-Hard 27.4% #30 /96
  • AraGen 30% #31 /42
Microsoft Apr 2024

phi-3-small 7.4B

Open weights

116.5 #192 of 256
Epoch Capabilities Index
Self-Reported Index 115.7 #363 of 404
  • SIQA 79.2% #2 /24
  • MHPP 19.5% #18 /23
  • MMLU 75.7% #93 /131
Microsoft Apr 2024

phi-3-medium 14B

Open weights Small

114.3 #193 of 256
Epoch Capabilities Index
Self-Reported Index 120.3 #345 of 404
  • GPQA Diamond 27.6% #213 /229
  • Open Arabic LLM Leaderboard v1 46.3% #21 /55
  • ChemBench 47.5% #19 /24
Microsoft Apr 2024

phi-3-mini 3.8B

Open weights

110.6 #203 of 256
Epoch Capabilities Index
Self-Reported Index 111.8 #371 of 404
  • SIQA 76.6% #4 /24
  • MHPP 19.5% #17 /23
  • CyberMetric 4.80% #10 /10
Microsoft Dec 2023

Phi-2

Open weights Small

96.4 #223 of 256
Epoch Capabilities Index
Self-Reported Index 104.5 #387 of 404
  • CyberMetric 52.1% #7 /10
  • MMLU 56.7% #121 /131
  • BIG-Bench Hard 59.2% #60 /73
Microsoft Sep 2023

Phi-1.5

Open weights

68.7 #248 of 256
Epoch Capabilities Index

What benchmarks does Microsoft mention most?

Here are the 28 benchmarks Microsoft mentions most in its launch posts.

Source: Labs’ launch posts, read by Benchmark Almanac

Model cards

DateModelDocumentBenchmarks
Oct 30, 2026 Mage-VL-Cowork Mage-VL-Cowork model card (Hugging Face) 1
Jul 27, 2026 MAI-Cyber-1-Flash Introducing MAI-Cyber-1-Flash inside MDASH (Microsoft AI news) 2
Jun 2, 2026 MAI-Code-1-Flash MAI-Code-1-Flash model card (Microsoft AI, PDF) 15
Jun 2, 2026 MAI-Thinking-1 MAI-Thinking-1: Building a Hill-Climbing Machine (Microsoft AI technical report), Tables 11 and 12 19
May 21, 2026 Fara1.5-27B Fara1.5-27B model card (Hugging Face) 3
Mar 4, 2026 Phi-4-Reasoning-Vision-15B Phi-4-Reasoning-Vision-15B model card (Hugging Face) 10
Nov 24, 2025 Fara-7B Fara-7B model card (Hugging Face) 4
Jul 9, 2025 Phi-4-mini-flash-reasoning Phi-4-mini-flash-reasoning model card (Hugging Face) 4
Apr 30, 2025 Phi-4-reasoning Phi-4-reasoning model card (Hugging Face) 12
Apr 30, 2025 Phi-4-mini-reasoning Phi-4-mini-reasoning model card (Hugging Face) 2
Apr 30, 2025 Phi-4-Reasoning-plus Phi-4-reasoning-plus model card (Hugging Face) 12
Apr 15, 2025 BitNet b1.58 2B4T BitNet b1.58 2B4T model card (Hugging Face) 16
Feb 26, 2025 Phi-4-multimodal Phi-4-multimodal-instruct model card (Hugging Face) 15
Feb 26, 2025 Phi-4 Mini Phi-4-mini-instruct model card (Hugging Face) 17
Dec 12, 2024 Phi-4 Phi-4 Technical Report (arXiv 2412.08905), Table 1 12
Aug 20, 2024 Phi-3.5-MoE Phi-3.5-MoE-instruct model card (Hugging Face), main benchmark table 21
Apr 22, 2024 phi-3-medium 14B Phi-3 Technical Report (arXiv 2404.14219v1), main benchmark table 21
Dec 12, 2023 Phi-2 Phi-2: The surprising power of small language models (Microsoft Research blog), Tables 1 and 2 14

Every Microsoft model