March 2026 monthly summary focused on delivering scalable LLM serving improvements in Blaizzy/mlx-vlm. Key achievement: integrated Qwen3-Omni-30B-A3B model with mlx_vlm server, including configuration, end-of-sequence handling, and exposing the language model property to the server for reliable text generation. Addressed integration reliability with a dedicated fix for Qwen3-Omni (qwen3_omni_moe) (#820). Completed end-to-end testing on target hardware configurations, validating deployment and performance. Impact: enables production-grade text generation via mlx_vlm, reduces operational risk, and positions the platform for higher throughput workloads. Technologies: LLM integration, model configuration, token handling, server exposure, testing across hardware, version control.