What's new in Foundry Labs - April 2026

2026-04-08 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Microsoft Foundry Labs has released several new AI models in April 2026, making advanced research prototypes accessible to developers. These include MAI-Transcribe-1, a speech recognition model offering 3.9% Word Error Rate on FLEURS and 50% lower GPU cost; MAI-Voice-1, a high-fidelity speech generation model; and MAI-Image-2, a text-to-image model ranking #3 on Arena.ai with 2x faster generation. Additionally, Foundry Labs introduced Harrier-oss-v1, an open-source multilingual text embedding model family supporting 94 languages, and Phi-4-Reasoning-Vision-15B, a 15B-parameter vision model for task-aware reasoning. VibeVoice ASR, a Microsoft Research model, provides long-form, structured speech recognition with speaker diarization and timestamping. Finally, GigaTIME, a multimodal AI model, translates H&E pathology slides into virtual mIF images across 21 protein channels, enabling population-scale tumor immune microenvironment modeling.

Key takeaway

For MLOps Engineers and AI developers building next-generation applications, exploring Microsoft Foundry Labs offers early access to high-performance, specialized AI models. You should evaluate MAI models for core speech and vision tasks, Harrier-oss-v1 for multilingual embeddings, and Phi-4-Reasoning-Vision-15B for vision reasoning to gain a competitive edge in your deployments.

Key insights

Microsoft Foundry Labs provides early access to advanced AI models for developers to explore and integrate.

Principles

• Cost-efficiency in AI inference is a key design goal.
• Multimodal and multilingual capabilities enhance AI utility.
• Smaller models can achieve competitive performance through distillation.

Method

GigaTIME uses a multimodal AI model trained on paired H&E and mIF data to generate virtual mIF images from low-cost H&E slides, enabling population-scale spatial proteomics studies.

In practice

• Use MAI-Transcribe-1 for cost-effective, accurate speech recognition.
• Deploy Harrier-oss-v1 for multilingual RAG pipelines.
• Integrate Phi-4-Reasoning-Vision-15B for visual reasoning in agentic applications.

Topics

• Microsoft Foundry Labs
• Speech AI
• Text-to-Image Generation
• Multilingual Text Embeddings
• Vision Reasoning

Best for: MLOps Engineer, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• Foundry Labs @ Build 2026 — Microsoft's Foundry Labs centralizes AI experiments, offering new models and community features across six key impact domains.
• Now in Foundry: Qwen3-Coder-Next, Qwen3-ASR-1.7B, Z-Image — New Hugging Face models offer efficient, high-performance AI for coding, multilingual ASR, and advanced image generation.
• Microsoft's AI Futurist explains how he uses Copilot — and the real-world problems enterprises are solving with agents — Microsoft's agent strategy centers on providing a comprehensive platform for context, governance, identity, and secure data access, alongside diverse model choices.
• Introducing Fireworks AI on Microsoft Foundry: Bringing high performance, low latency open model inference to Azure — Fireworks AI on Microsoft Foundry provides high-performance, enterprise-grade open model inference within Azure's unified AI platform.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.