Microsoft’s First Homegrown Models Are Here
Good morning AI entrepreneurs & enthusiasts,
For years, Microsoft's AI playbook was tightly interwoven with OpenAI. That just changed.
With the unveiling of MAI-Voice-1 and MAI-1-preview, Microsoft has stepped into the spotlight with its first internally built models - a pivotal move that could reshape one of the industry’s most high-profile alliances.
Today in AI news:
Microsoft rolls out in-house AI models
gpt-realtime: Real-time speech AI goes mainstream
Grok Code Fast 1: High-speed agentic coding assistant
Codex gets major upgrade: IDE, CLI, and GitHub reviews
TIME100 AI 2025: The most influential voices in AI
Top Tools & Quick News
MICROSOFT 🤖 Microsoft launches first internal AI models
News: Microsoft just released MAI-Voice-1 and MAI-1-preview—its first fully homegrown models, following years of dependency on OpenAI.
Details:
MAI-Voice-1: Produces 1 min of expressive audio in <1 sec on a single GPU, supporting multi-speaker styles. Already integrated into Copilot Daily + Podcasts, with high-fidelity natural speech.
MAI-1-preview: Trained on ~15,000 Nvidia H100 GPUs, optimized for instruction-following and general tasks, currently ranked 13th on LMArena benchmarks.
Testing is underway via limited API access and public preview deployments.
Why it matters: This launch represents a bid for independence, embedding proprietary models directly into Windows, Office, Teams, and Copilot. Supported by multibillion-dollar infrastructure investments, the move reduces reliance on OpenAI while cementing Microsoft’s positioning across cloud, productivity, and AI platforms. While the OpenAI partnership remains in place through 2030, these in-house models clearly mark the beginning of a parallel path toward autonomy and long-term AI leadership.
XAI ⚡ Grok Code Fast 1: High-speed agentic coding assistant
News: xAI has launched Grok Code Fast 1, a next-gen coding model built for high-speed agentic coding with enterprise-level performance.
Details:
Built on a 314B parameter Mixture-of-Experts architecture with a 256k token context window
Generates ~160–190 tokens/second, over 60% faster than rivals like Gemini 2.5 Pro
Pricing: $0.20 per million input, $1.50 per million output, $0.02 per million cached tokens
Supports languages including TypeScript, Python, Java, Rust, C++, and Go
Offers function/tool call support and visible reasoning traces
API integrations with GitHub Copilot, Cursor, Roo Code, Windsurf, and others
Launch partners providing free preview access; see full model card for specs
Why it matters: By combining agentic autonomy, blazing speed, and ultra-low costs, xAI is positioning Grok Code Fast 1 as a daily driver for real-world coding. For developers, it means faster iterations, scalable workflows, and practical usability that undercuts rivals on both performance and price.
OPENAI 💻 Codex gets major upgrade: IDE, CLI, and GitHub reviews
News: OpenAI has rolled out a major Codex update, adding deep integrations for IDEs, CLI, and GitHub workflows.
Details:
IDE Extension: Works in VS Code + compatible editors, supports local file editing and async cloud offload without losing state.
Codex CLI: Refreshed UI, new commands, sign-in with ChatGPT accounts (no API key required).
Code Reviews: Auto-reviews GitHub PRs or responds to mentions with suggested fixes.
Seamless Handoff: Start locally, offload tasks to the cloud without losing state.
Model backbone: Powered by GPT-5, OpenAI’s latest coding model.
Why it matters: Codex has evolved from a niche autocomplete tool into a full-stack AI software engineering partner. Deeply embedded in editors, terminals, and GitHub, it can now generate, test, and review code across the entire development lifecycle. By reducing friction and automating repetitive tasks, OpenAI positions Codex as an indispensable co-pilot that accelerates iteration, enforces quality, and transforms how software is built and maintained.
TIME 📰 TIME100 AI 2025: The most influential voices in AI
News: TIME has released the third edition of its TIME100 AI list, spotlighting 100 people shaping the future of artificial intelligence.
Details:
Highlights the talent wars: firms like Meta offering nine-figure deals to attract stars such as Alexandr Wang and Nat Friedman.
Features diverse voices: Pope Leo XIV and Yoshua Bengio cautioning on AI’s risks.
Artist Refik Anadol designed this year’s cover using AI trained on historic TIME covers.
Notable honorees include Sam Altman (OpenAI), Elon Musk (xAI), Jensen Huang (NVIDIA), and Mark Zuckerberg (Meta).
Why it matters: The TIME100 AI list reflects the cultural and industrial weight of AI today—from billion-dollar recruiting battles to creative expression and societal concerns. It underscores how AI leaders and critics alike are shaping the next chapter of technology’s integration into everyday life.
🛠 Trending Tools:
Google Vids: AI-powered video creation/editing
MAI-Voice-1: Microsoft’s speech model
gpt-realtime: OpenAI’s new real-time agent
HunyuanVideo-Foley: SOTA open-source audio generator
🔍 Quick News:
Anthropic exposes Claude Code abuse by cybercriminals, who used it to automate large-scale extortion attacks, a technique termed “vibe hacking.”
Krea previews Realtime Video, offering paint-on-canvas editing for dynamic video creation.
Tencent drops HunyuanVideo-Foley, an open-source tool for professional-grade synchronized sound effects.
Thanks for reading this far! Stay ahead of the curve with my daily AI newsletter—bringing you the latest in AI news, innovation, and leadership every single day, 365 days a year. See you tomorrow for more!