01 What happened
Microsoft Foundry Managed Fine-Tuning expands its collection with new open-source models, the first Microsoft model for fine-tuning (MAI-Code-1.1-Flash), and Multi-LoRA shared GPU deployments, all now in public preview.
02 Key details
- Foundry's managed fine-tuning collection gains Muse-Glimmer-30B, qwen3.8-27b, qwen3.6-35b-a3b, and gpt-oss-120b in public preview, each supporting SFT and RFT.
- MAI-Code-1.1-Flash, the first Microsoft model for Foundry fine-tuning, is available in public preview with RFT-only support.
- Multi-LoRA Managed Compute lets compatible adapters share base-model weights and GPU capacity, billed hourly instead of per-token inference.
03 Why it matters
These updates reduce hosting costs for multiple fine-tuned variants via shared hardware and expand the range of customizable models, including the first Microsoft coding model for fine-tuning in Azure AI Foundry.
04 Who it matters to
AI developers and machine learning engineers.
Original source