Trending Now - Microsoft Unveils Three New AI Models to Compete With Rivals

WHO:

CNBC - Microsoft Vice Chair and President Brad Smith met with Japan's Prime Minister Sanae Takaichi on April 3rd, 2026.

CNBC - Microsoft Vice Chair and President Brad Smith met with Japan's Prime Minister Sanae Takaichi.

VentureBeat - Mustafa Suleyman stated, "We’re now a top three lab just under OpenAI and Gemini."

VentureBeat - Mustafa Suleyman stated, "I’m very excited that we’ve now got the first models out, which are the very best in the world for transcription."

VentureBeat - Suleyman emphasized that the audio model was built by a team of 10 engineers and the image team was also less than 10 people.

Techmeme - Microsoft, led by its superintelligence team, which was formed by Suleyman.

VentureBeat - Suleyman, Microsoft’s superintelligence team leader.

WHAT:

CNBC - Microsoft plans to invest $10 billion in Japan between 2026 and 2029 to build AI infrastructure, strengthen cybersecurity, and train 1 million engineers and developers by 2030.

CNBC - Shares of Sakura Internet surged as much as 20.2% following Microsoft's announcement of the investment.

CNBC - Microsoft plans to invest $10 billion in Japan between 2026 and 2029 to build AI infrastructure, strengthen cybersecurity, and train 1 million engineers and developers by 2030.

CNBC - Shares of Sakura Internet surged as much as 20.2% following Microsoft's announcement of the investment.

VentureBeat - MAI-Transcribe-1 beats ElevenLabs' Scribe v2 and OpenAI's GPT-Transcribe on 15 of 25 languages.

VentureBeat - MAI-Voice-1 generates audio at 60 times real-time speed.

VentureBeat - Microsoft is testing MAI-Transcribe-1 inside Copilot's Voice mode and Microsoft Teams for conversation transcription.

VentureBeat - Microsoft’s MAI-Transcribe-1 model claims to achieve a 3.8% Word Error Rate on the FLEURS benchmark across 25 languages, outperforming OpenAI's Whisper-large-v3 on all 25 languages and Google’s Gemini 3.1 Flash on 22 of 25 languages.

VentureBeat - MAI-Voice-1 generates 60 seconds of audio in one second and is priced at $22 per 1 million characters.

VentureBeat - MAI-Image-2 debuted as a top-three model family on the Arena.ai leaderboard and delivers at least 2x faster generation times compared to its predecessor.

VentureBeat - Microsoft is rolling out MAI-Image-2 across Bing and PowerPoint, pricing it at $5 per 1 million tokens for text input and $33 per 1 million tokens for image output.

VentureBeat - WPP, one of the world’s largest advertising holding companies, is among the first enterprise partners building with MAI-Image-2 at scale.

VentureBeat - MAI-Transcribe-1 uses a transformer-based text decoder with a bi-directional audio encoder.

VentureBeat - MAI-Voice-1 preserves speaker identity across long-form content and supports custom voice creation from just a few seconds of audio.

VentureBeat - MAI-Image-2 is now rolling out across Bing and PowerPoint.

VentureBeat - Microsoft’s MAI-Transcribe-1 model claims to achieve the lowest average Word Error Rate on the FLEURS benchmark across the top 25 languages, outperforming OpenAI's Whisper-large-v3 on all 25 languages and Google’s Gemini 3.1 Flash on 22 of 25 languages.

VentureBeat - MAI-Transcribe-1 accepts MP3, WAV, and FLAC files up to 200MB and has a batch transcription speed that is 2.5 times faster than the existing Microsoft Azure Fast offering.

VentureBeat - MAI-Voice-1 supports custom voice creation from just a few seconds of audio through Microsoft Foundry.

VentureBeat - MAI-Image-2 debuted as a top-three model family on the Arena.ai leaderboard and delivers at least 2x faster generation times compared to its predecessor.

Techmeme - Microsoft launched three new AI models: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2.

VentureBeat - The models are designed for speech transcription, voice generation, and image creation, respectively.

VentureBeat - MAI-Transcribe-1 achieves a 3.8% Word Error Rate on the FLEURS benchmark across 25 languages.

VentureBeat - MAI-Voice-1 generates 60 seconds of audio in one second and is priced at $22 per 1 million characters.

VentureBeat - MAI-Image-2 is priced at $5 per 1 million tokens for text input and $33 per 1 million tokens for image output.

WHERE:

CNBC - The investment will focus on developing AI infrastructure in Japan, specifically utilizing Sakura Internet's domestic data centers.

VentureBeat - The models are available through Microsoft Foundry and a new MAI Playground.

WHEN:

CNBC - The announcement was made on April 3rd, 2026.

VentureBeat - Microsoft’s superintelligence team was formally established in October 2025.

VentureBeat - The models were launched on March 31st, 2026.

VentureBeat - Suleyman mentioned that Microsoft’s superintelligence team was formally established in October 2025.

Techmeme - The models were launched on Wednesday, March 31st, 2026.

VentureBeat - The superintelligence team was formed six months prior to the launch, in September 2025.

VentureBeat - Microsoft renegotiated its contract with OpenAI in October 2025.

WHY:

CNBC - The investment comes as demand for cloud and AI services grows in Japan, where around one in five working-age people use generative AI tools, compared to the global average of about one in six, according to Microsoft's AI Diffusion Report.

CNBC - The investment comes as demand for cloud and AI services grows in Japan, where around one in five working-age people use generative AI tools.

VentureBeat - The launch is a response to investor pressure for proof that AI infrastructure spending will translate into revenue, as Microsoft’s stock has fallen roughly 17% year-to-date.

VentureBeat - The launch is a response to investor pressure for proof of revenue from AI infrastructure spending, as Microsoft’s stock has fallen roughly 17% year-to-date.

VentureBeat - Suleyman described the goal of the superintelligence team as focusing all energy on delivering world-class models for Microsoft over the next 5 years.

VentureBeat - Microsoft’s pricing strategy for MAI-Voice-1 and MAI-Image-2 is designed to be the cheapest among major hyperscalers, including Amazon and Google.

VentureBeat - Microsoft aims to compete directly with OpenAI, Google, and other AI labs on model development.

VentureBeat - The launch is a response to investor pressure for proof of revenue from AI infrastructure spending.

HOW:

CNBC - Microsoft will partner with Sakura Internet and SoftBank to provide AI computing resources, including graphics processing units located in Japan.

VentureBeat - Suleyman described the working environment of his team as resembling a startup trading floor, with groups of people around circular tables, emphasizing a flat organizational structure.

VentureBeat - The audio model was built by a team of 10 engineers, emphasizing a flat organizational structure that promotes empowerment and efficiency.

VentureBeat - The models were built by small teams, with the audio model developed by 10 engineers.

VentureBeat - The models utilize advanced architecture and data innovation to achieve high performance.

VentureBeat - Microsoft’s models run on half the GPUs compared to competitors, enhancing efficiency and reducing costs.

(UPDATED April 3, 2026 13:18 UTC)

on Friday, Microsoft Vice Chair and President Brad Smith met with Japan's Prime Minister Sanae Takaichi to announce a significant investment of $10 billion in the country from 2026 to 2029. This funding aims to develop AI infrastructure, enhance cybersecurity, and train one million engineers and developers by 2030, leading to a surge in shares of Sakura Internet by over 20 percent. The investment is driven by the growing demand for cloud and AI services in Japan, where approximately one in five working-age individuals are using generative AI tools.

(UPDATED April 3, 2026 07:17 UTC)

on Friday, Microsoft announced plans to invest $10 billion in Japan over the next few years to build AI infrastructure, enhance cybersecurity, and train one million engineers and developers by 2030. This investment, which will focus on utilizing Sakura Internet's domestic data centers, has already led to a significant surge in Sakura Internet's shares, climbing as much as 20.2%. Microsoft will collaborate with Sakura Internet and SoftBank to provide essential AI computing resources, including graphics processing units based in Japan, in response to the growing demand for cloud and AI services in the country.

(UPDATED April 3, 2026 03:50 UTC)

Microsoft's superintelligence team, established in October 20th25, is making waves in the AI landscape, with Mustafa Suleyman announcing that they are now a top three lab just under OpenAI and Gemini. Their new MAI-Transcribe-1 has outperformed ElevenLabs' Scribe v2 and OpenAI's GPT-Transcribe in 15 out of 25 languages, while MAI-Voice-1 can generate audio at an impressive 60 times real-time speed. This launch comes as a response to investor pressure for tangible results, especially as Microsoft’s stock has seen a decline of about 17% year-to-date, with the team’s working environment designed to foster innovation through a flat organizational structure.

(UPDATED April 2, 2026 19:18 UTC)

Microsoft has launched a new suite of AI models, including MAI-Transcribe-1, which boasts a 3.8% Word Error Rate on the FLEURS benchmark across 25 languages, outperforming competitors like OpenAI's Whisper-large-v3 and Google’s Gemini 3.1 Flash. The MAI-Voice-1 model can generate 60 seconds of audio in just one second, while MAI-Image-2, which has debuted as a top-three model on the Arena.ai leaderboard, is being rolled out across Bing and PowerPoint. This launch comes as a response to investor pressure for revenue proof amid a 17% drop in Microsoft’s stock this year, and the team behind these innovations operates in a startup-like environment with a flat organizational structure.

(UPDATED April 2, 2026 17:16 UTC)

Mustafa Suleyman announced that his team has released the first models for transcription, which he claims are the best in the world. The MAI-Transcribe-1 model utilizes a transformer-based text decoder and a bi-directional audio encoder, while MAI-Voice-1 can maintain speaker identity in long-form content and create custom voices from just a few seconds of audio. Suleyman highlighted that the Microsoft superintelligence team, established in October 20th25, operates in a startup-like environment with a flat organizational structure to focus on delivering world-class models over the next five years.

(UPDATED April 2, 2026 15:18 UTC)

Microsoft has announced its MAI-Transcribe-1 model, which boasts the lowest average Word Error Rate on the FLEURS benchmark across the top 25 languages, surpassing OpenAI's Whisper-large-v3 and Google’s Gemini 3.1 Flash in performance. The model accepts various audio file formats and offers a transcription speed that is two and a half times faster than Microsoft Azure's existing Fast offering, while the MAI-Voice-1 allows for custom voice creation from just a few seconds of audio. Additionally, Microsoft is positioning the pricing for MAI-Voice-1 and MAI-Image-2 to be the most competitive among major hyperscalers, with the audio model developed by a team of 10 engineers focused on a flat organizational structure for enhanced efficiency.

(UPDATED April 2, 2026 13:21 UTC)

Microsoft has launched three new AI models designed for speech transcription, voice generation, and image creation, named MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, on Tuesday. The superintelligence team, led by Suleyman and formed six months prior, developed these models, which boast impressive performance metrics, including MAI-Transcribe-1 achieving a 3.8% Word Error Rate across 25 languages and MAI-Voice-1 generating 60 seconds of audio in just one second. These models are now available through Microsoft Foundry and a new MAI Playground, as the company aims to compete directly with OpenAI and Google amid increasing investor pressure for AI revenue proof.