Now in Foundry: Microsoft Harrier and NVIDIA EGM-8B

2026-04-13 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, medium

Summary

Microsoft Foundry now supports two new efficient AI models: Microsoft Research's harrier-oss-v1-0.6b and NVIDIA's EGM-8B. The harrier-oss-v1-0.6b, a 0.6B parameter decoder-only model, achieves a 69.0 score on the Multilingual MTEB v2 leaderboard for text embeddings, utilizing contrastive learning and knowledge distillation. It supports over 100 languages and handles six embedding scenarios with task-instruction queries. NVIDIA's EGM-8B, an ~8.8B parameter Vision Language Model, scores 91.4 average IoU on the RefCOCO visual grounding benchmark, showing a +3.6 IoU gain over its base model through two-stage training with Supervised Fine-Tuning and Group Relative Policy Optimization. Both models demonstrate that targeted training strategies can achieve performance comparable to much larger models, emphasizing efficiency-first development.

Key takeaway

For AI Architects and NLP Engineers evaluating model deployment, consider harrier-oss-v1-0.6b for multilingual embedding tasks and EGM-8B for visual grounding. These models offer high performance at reduced parameter counts, allowing for more efficient resource utilization and faster inference, which can significantly lower operational costs and improve latency in production environments. Explore their one-click deployment options in Microsoft Foundry.

Key insights

Targeted training strategies enable smaller models to achieve performance comparable to larger, more resource-intensive counterparts.

Principles

• Efficiency-first model development narrows the small-large model gap.
• Decoder-only architecture can excel in embedding tasks.
• Reinforcement Learning improves visual grounding accuracy.

Method

harrier-oss-v1-0.6b uses contrastive learning and knowledge distillation. EGM-8B employs two-stage training: SFT on chain-of-thought traces followed by GRPO with an IoU/task success reward.

In practice

• Use harrier-oss-v1-0.6b for multilingual semantic search.
• Apply EGM-8B for object localization in images.
• Deploy models via Microsoft Foundry for scalable inference.

Topics

• Microsoft Harrier
• NVIDIA EGM-8B
• Multilingual MTEB v2
• Visual Grounding
• Knowledge Distillation

Best for: AI Architect, NLP Engineer, Computer Vision Engineer, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Lambda's MLPerf Inference v6.0: hardware leap, software maturity, research breakthrough — Hardware, software, and routing optimizations significantly boost AI inference performance and reduce latency.
• Thinking with Visual Grounding — Vision-language models enhance reasoning by explicitly grounding natural-language thoughts to specific image regions.
• 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.
• NVIDIA Drops a Model “LocateAnything” — NVIDIA's LocateAnything model uses Parallel Box Decoding for rapid visual grounding, enabling efficient agent AI workflows.
• Training the AIs' Eyes: How Roboflow is Making the Real World Programmable, with CEO Joseph Nelson — Computer vision, though trailing language models, is approaching its "ChatGPT moment" through specialized, efficient, and edge-deployable models.
• UniversalVTG: A Universal and Lightweight Foundation Model for Video Temporal Grounding — UniversalVTG offers a lightweight, cross-dataset pretraining approach for video temporal grounding, outperforming larger MLLMs.

Open in AIssential →

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