Microsoft Foundry adds open-model fine-tuning, MAI-Code RFT and shared multi-LoRA deployments
Foundry managed fine-tuning now covers four open models, including Qwen3.6-35B-A3B and gpt-oss-120b, with both supervised and reinforcement fine-tuning. MAI-Code-1.1-Flash can be customized with RFT only, and a new multi-LoRA managed compute option lets compatible adapters share one base-model deployment. Everything is in public preview with hourly hardware billing.
Why it matters
Sharing one base deployment across many LoRA adapters can sharply cut the cost of serving per-team or per-customer fine-tunes on Azure.