MagenticBrain

MagenticBrain is an 8-billion-parameter orchestration model from Microsoft Research AI Frontiers, built as the brain of the MagenticLite agent stack. Fine-tuned from Qwen 3 on trajectories that blend tool-calling with coding and terminal sequences, it turns vague natural-language requests into concrete plans, picks the right tool or sub-agent for each step, writes code when code is the right answer, and recovers when something breaks mid-task. Explicit delegation training teaches the orchestrator when to hand a subtask off — most notably routing browser-based work to Fara1.5, the computer-use specialist in the same stack.

MagenticBrain is trained end-to-end inside the MagenticLite harness using the same tool schemas and execution environment it sees at inference time, closing the gap between training-time orchestration patterns and deployment behavior that bedevils most agent systems. At 8B parameters, it brings serious planning and coding capability into the range where multi-agent applications and on-device assistants stay fast and cheap — places frontier orchestrators are too expensive to deploy. Together with Fara1.5, it completes the recipe for end-to-end agentic apps that run efficiently on commodity hardware.