Microsoft released three in-house AI models on April 2 — MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — built entirely by its own superintelligence team and available immediately through Microsoft Foundry. The models handle speech-to-text transcription, voice generation, and image creation, directly competing with OpenAI's Whisper, text-to-speech, and DALL-E offerings. Microsoft priced them aggressively below every major cloud competitor.
This matters well beyond a corporate rivalry. When the company that has invested $13 billion into OpenAI starts building and selling its own competing models at cut-rate prices, it accelerates a pricing war that makes AI tools meaningfully cheaper for every business. If you're running operations that touch transcription, voice agents, or image generation — and an increasing number of businesses are — your costs just dropped.
What Microsoft actually shipped
The three models each target a commercially valuable capability that enterprises are already spending on.
MAI-Transcribe-1 delivers speech-to-text across 25 languages with what Microsoft claims is the lowest average Word Error Rate on the FLEURS benchmark — the industry-standard multilingual test — at 3.8% WER. According to Microsoft's benchmarks, it beats OpenAI's Whisper-large-v3 on all 25 languages and Google's Gemini 3.1 Flash on 22 of 25. Batch transcription runs 2.5 times faster than Microsoft's own previous Azure Fast offering, at roughly 50% lower GPU cost. It starts at $0.36 per hour.
MAI-Voice-1 generates natural speech from text, producing 60 seconds of audio in a single second. It supports custom voice creation from just a few seconds of sample audio — a capability that puts it in direct competition with ElevenLabs and Resemble AI. Pricing starts at $22 per million characters.
MAI-Image-2 debuted as a top-three model family on the Arena.ai leaderboard and delivers at least 2x faster generation times. WPP, one of the world's largest advertising groups, is among the first enterprise partners building with it at scale. "MAI-Image-2 is a genuine game-changer," said Rob Reilly, Global Chief Creative Officer at WPP. "It's a platform that not only responds to the intricate nuance of creative direction, but deeply respects the sheer craft involved in generating real-world, campaign-ready images." It starts at $5 per million tokens for text input and $33 per million tokens for image output.
All three models are already powering Microsoft's own products — Copilot, Bing, PowerPoint, and Azure Speech — which is Microsoft's way of saying they've been battle-tested in production before reaching third-party developers.
The strategic break from OpenAI
The models were built by Microsoft's MAI Superintelligence team, formed in November 2025 under Mustafa Suleyman, CEO of Microsoft AI. Their existence is a direct consequence of a contractual renegotiation that, until October 2025, prohibited Microsoft from independently pursuing AGI.
"Back in September of last year, we renegotiated the contract with OpenAI, and that enabled us to independently pursue our own superintelligence," Suleyman told VentureBeat. "Since then, we've been convening the compute and the team and buying up the data that we need."
The timing is pointed. Microsoft's stock just closed its worst quarter since the 2008 financial crisis, down roughly 17% year-to-date, as investors demand proof that hundreds of billions in AI infrastructure spending will produce returns. Meanwhile, OpenAI is projected to lose $14 billion this year, according to internal projections published by The Information. Microsoft's decision to build models that run on half the GPUs of competitors isn't just about competing with OpenAI — it's about proving to shareholders that AI can be an efficient business, not a cash bonfire.
Suleyman was careful to note the OpenAI partnership remains intact through at least 2032. But his stated ambition is unmistakable: "Our mission is to make sure that if Microsoft ever needs it, we will be able to provide state of the art at the best efficiency, the cheapest price, and be completely independent," he told VentureBeat.
Ten people built a world-class model
Perhaps the most striking detail is scale — or lack of it. Suleyman revealed that the audio model was built by a team of just 10 engineers, and the image team was similarly small. "The vast majority of the speed, efficiency and accuracy gains come from the model architecture and the data that we have used," he said. "My philosophy has always been that we need fewer people who are more empowered."
This challenges the prevailing narrative that frontier AI development requires thousands of researchers and billions in headcount. If state-of-the-art transcription can be built by 10 people using half the industry-standard GPU footprint, the economics of the AI industry look very different from the narrative that has justified massive spending.
Why this matters for your business
Here's where this gets practical. AI API prices have dropped 40-70% across the board in early 2026, driven by compute cost reductions, market competition, and challengers like DeepSeek offering free capabilities. Microsoft entering with aggressively priced, enterprise-grade models accelerates that trend.
If you're a business running any of these workloads, the calculus has shifted:
Transcription and meeting intelligence. If you're paying for meeting transcription, call centre analytics, or multilingual content processing, MAI-Transcribe-1 at $0.36 per hour — with claimed best-in-class accuracy across 25 languages — changes your cost equation. For a 50-person company running daily meetings, that's a material reduction in tooling spend.
Voice agents and IVR. Custom voice creation from seconds of audio, at $22 per million characters, makes branded voice experiences accessible to mid-market businesses. If you've been evaluating voice AI for customer service, the barriers just got lower.
Content and creative. Image generation at $5 per million input tokens, integrated into the same platform you likely already use for other AI services, reduces the friction of adding AI-generated visuals to marketing workflows.
The bigger picture: competition between Microsoft, Google, OpenAI, and Anthropic is compressing margins and improving quality simultaneously. The CostLayer analysis of Q1 2026 pricing found that major providers have cut prices by an average of 67% over the past year. That trend shows no sign of slowing down.
What to watch
Suleyman confirmed that a frontier large language model is coming — a direct competitor to GPT and Gemini at the reasoning and text generation level. That would be a fundamentally different proposition from the specialised models launched this week, and would represent the real test of whether Microsoft can stand on its own as an AI lab.
The key milestone to monitor: whether Microsoft ships a competitive LLM before its investors' patience runs out, and whether OpenAI responds by adjusting the terms of a partnership that increasingly looks like it benefits Microsoft more than it costs. For business owners, the implication is clear — lock in nothing, stay flexible, and let the giants fight over your dollar. The pricing war has barely started.
Sources
- Today we're announcing 3 new world class MAI models, available in Foundry — Microsoft AI (Mustafa Suleyman)
- Microsoft launches 3 new AI models in direct shot at OpenAI and Google — VentureBeat
- Microsoft takes on AI rivals with three new foundational models — TechCrunch
- Microsoft shivs OpenAI with three new AI models for speech and images — The Register
- Microsoft released 3 new AI models, ramping up competition with OpenAI — Business Insider
- 2026 AI Pricing War: How 67% Cost Cuts Are Reshaping LLM Costs — CostLayer