Microsoft releases 4B vision-language model Mage-VL-Cowork

Microsoft released Mage-VL-Cowork on Hugging Face, a 4B vision-language model for autonomous desktop agents. Combining teacher-harness evolution, the Mage-CUA-Forge workflow environment synthesis and a five-stage curriculum, it scores 66.0% on OSWorld-Verified while cutting model calls 13.0% and output tokens 25.4% after online RL.

• Reports of OpenAI model misalignment grow, prompting calls for stronger guardrails2026-10-14
• ThinkingBox benchmark: one-shot agent success rarely repeats2026-10-09
• Microsoft releases ThinkingBox to grade AI agents on database state2026-10-03
• AI Paradox Needs Guardrails, Not Panic2026-09-26
• Expert: AI training does not make online content free to use2026-10-31