Organization
Microsoft
Tracked models
78
Frontier models
2
Largest training run
7 x 10^24 FLOP
Data-center compute
1.7M H100e
78 tracked AI models list Microsoft as a developer. 2 of them are frontier models — among the largest training runs of their moment. The largest is MAI-Thinking-1, at 7 x 10^24 FLOP (June 2026). The most recent tracked release is from June 2026.
It owns 7 tracked AI data centers, holding about 1.7M H100e and drawing 1.52 GW. It operates 15 tracked GPU clusters.
Models (78)
MAI-Thinking-17 x 10^24 FLOPPhi-4-Reasoning9.3 x 10^23 FLOPMegatron-Turing NLG 530B8.6 x 10^23 FLOPphi-3-medium 14B4 x 10^23 FLOPPhi-3.5-MoE3 x 10^23 FLOPphi-3-small 7.4B2.1 x 10^23 FLOPPhi-4-Multimodal1.2 x 10^23 FLOPWizardCoder-15.5B1.1 x 10^23 FLOPPhi-4 Mini10 x 10^22 FLOPphi-3.5-Vision8.8 x 10^22 FLOPphi-3-mini 3.8B7.5 x 10^22 FLOPSamba 3.8B7.3 x 10^22 FLOPFlorence4.8 x 10^22 FLOPWizardLM-7B4 x 10^22 FLOPphi-3.5-mini3.7 x 10^22 FLOPTuring ULRv52.9 x 10^22 FLOPDeBERTa2.6 x 10^22 FLOPPhi-22.3 x 10^22 FLOPMicrosoft MAI-1 (2024 unreleased)1.6 x 10^22 FLOPTuring-NLG1.6 x 10^22 FLOPGPT3-6.7B + muP1.3 x 10^22 FLOPFlorence-2-L (large)1.2 x 10^22 FLOPGPT3-6.7B (rerun of original)1.2 x 10^22 FLOPXY-LENTXL7.2 x 10^21 FLOPFlorence-2-B (base)5.4 x 10^21 FLOPLongNet4.9 x 10^21 FLOPKosmos-13.5 x 10^21 FLOPGPT-2 (1.5B, Curriculum Learning 45K)2.4 x 10^21 FLOPKosmos-2.52.2 x 10^21 FLOPVALL-E X1.2 x 10^21 FLOPPhi-1.51.2 x 10^21 FLOPE2 TTS4.9 x 10^20 FLOPKosmos-24.6 x 10^20 FLOPGPT-2 (117M, SLW 110K)1.1 x 10^20 FLOPBEIT-37 x 10^19 FLOPT2R + Random Init2.7 x 10^19 FLOPT2R + Pretrain1.4 x 10^19 FLOPT2R 75% + Pretrain (WT-103)1.4 x 10^19 FLOPResNet-152 (ImageNet)10^19 FLOPVALL-E10^19 FLOPCODEFUSION (Python)7.9 x 10^18 FLOPResNet-101 (ImageNet)7 x 10^18 FLOPSPPNet3.4 x 10^18 FLOPMultilingual-E5-large3.4 x 10^18 FLOPISS3.4 x 10^15 FLOPRankNet3.5 x 10^12 FLOPMAI-Image-2.5June 2026MAI-Image-1October 2025MAI-Voice-1August 2025CollabLLMJune 2025Phi-4-Reasoning-plusApril 2025MAI-DS-R1April 2025ProteinGeneratorSeptember 2024GRITLM 7BApril 2024GRITLM 8x7BApril 2024WizardLM-2 70BApril 2024WizardLM-2 8x22BApril 2024WizardLM-2 7BApril 2024MiniGPT4 + LRV-InstructionMarch 2024WizardLM 13B v1.2July 2023WizardLM 70BApril 2023Diffusion-GANJune 2022DeBERTaV3large + KEARMay 2022T-NLRv5 XXLDecember 2021Quantized ADMMNovember 2021MEBSeptember 2021HJRSSSeptember 2021Theseus 6/768February 2020MT-DNNJanuary 2019Chinese - English translationMarch 2018HRAJune 2017ResNet-1001September 2016MS-ensemble-speech-recognitionSeptember 2016ResNet-110 (CIFAR-10)December 2015DBN for NLPFebruary 2014RNN+LDA+KN5+cacheDecember 2012Stacked Semisuperviser AutoencodersJuly 2008MAI-Code-1-Flash
Data centers (7)
GPU clusters (15)
Microsoft Azure Eagle14.4K H100eMicrosoft Azure ND H100 v5 VM4K H100eMicrosoft Azure Meta AI Rental1.7K H100eMicrosoft Ares/Maia828 H100eMicrosoft Azure Voyager-EUS2666 H100eND A100 v4631 H100eMicrosoft Azure Pioneer-WUS2414 H100eMicrosoft Azure Pioneer-WEU414 H100eMicrosoft Azure Pioneer-EUS414 H100eMicrosoft Azure Pioneer-SCUS414 H100eMicrosoft Explorer-WUS3372 H100eMicrosoft Azure Immunity Bio78.95 H100eMicrosoft Azure ND v2 Largest Stated50.53 H100eMicrosoft Research Hypercluster36.38 H100eMicrosoft Air Gapped Supercomputer
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
Other organizations
xAIOpenAIMeta AIGoogle DeepMindNVIDIACursorAnthropicByteDanceMoonshotAlibabaHuaweiSaudi AramcoInflection AIBeijing Academy of Artificial Intelligence / BAAIDeepSeekReka AIAI SingaporeGoogleGoogle ResearchChina TelecomZ.ai (Zhipu AI)Xiaomi CorpETH ZurichEcole Polytechnique Federale de Lausanne (EPFL)