Microsoft is one of the early angel of OpenAI , and has repeatedly monger products like co-pilot by swash their approach to the latest ChatGPT models . Now , it seems Microsoft is looking to push its own AI exemplar in the pop software suite , while also developing a rival to OpenAI ’s logical thinking models in the ” GPT - o ” menage .
As perThe Information , employee at Microsoft ’s AI unit recently concluded the training of “ a new syndicate of AI model ” that are presently in development under the “ MAI ” codename . Internally , the team is promising that these in - house models do nearly as well as the top AI framework from the likes of OpenAI and Anthropic .
Under the leaders of its AI chief , Mustafa Suleyman , Microsoft is found this enterprisingness to trim down its dependence on OpenAI and develop its own AI flock forCopilot diligence . The developments are not surprising .
Steadily building its own stack
In the last hebdomad of February , Microsoft introduced new small language models telephone Phi-4 - multimodal and Phi-4 - mini . They occur with multi - modal capability , which mean they can process textbook , speech , and vision as input formats , just like OpenAI ’s ChatGPT andGoogle ’s Gemini .
These two new AI framework are already uncommitted to developer via Microsoft ’s Azure AI Foundry and third - party program such as HuggingFace and the NVIDIA API Catalog . In benchmarks shared by the company , the Phi-4 model is already ahead ofGoogle ’s latest Gemini 2.0 series modelson multiple test parametric quantity .
“ It is among a few loose models to successfully implement language summarization and achieve public presentation levels comparable to GPT-4o model , ” Microsoft mention in itsblog post . The company is hop-skip to release its ” MAI ” models commercially via its Azure military service .
Rivalry, and openness to rivals
Aside from testing in - house AI theoretical account for Copilot , Microsoft is also exploring third - party options such as DeepSeek , xAI , and Meta . DeepSeek recently made wavesby offering a high performance benchmark at a dramatically low development toll . It has already beenadopted by numerous companiesand recentlyclaimeda theoretical price - to - profit ratio of over 500 % on a daily basis .
Today , we are advancing our AI aspiration with the release of DeepSeek R1 7B & amp ; 14B distilled example for Copilot+ PC via Azure AI Foundry . This is the next step on our journey to keep to make Windows the platform for AI , seamlessly integrating intelligence from the swarm to…pic.twitter.com/QaUYrlMIt6
& mdash ; Pavan Davuluri ( @pavandavuluri)March 3 , 2025
Aside from grow its own AI model to put back OpenAI ’s GPT infrastructure for Copilot , Microsoft is also reportedly working on its own logical thinking AI models , too . This would pit Microsoft againstOpenAI products such as GPT - o1as well as Chinese upstarts such as DeepSeek , both of which offer up reasoning capabilities .
obviously , the work on an in - house reasoning model has been expedited due to strained kinship between Microsoft and OpenAI squad over technology sharing . According to The Information , Suleyman and OpenAI have been at odds over the latter ’s lack of transparency regarding the intricate workings of its AI models such as GPT - o1 .
Reasoning models are hold to be the next frontier for AI development , as they declare oneself a more nuanced understanding of question , logical deduction , and serious job solve capabilities . Microsoft also claims that its Phi-4 mannequin deliver stronger language , mathematical , and ocular science reasoning chops .
