News › ai-infrastructure
By Zayden R., August 12, 2026
Intel's LLM-Scaler project now supports Muse Glimmer 30B with FP8 quantization, adding new AI capabilities to Arc Pro B-Series cards. This update also includes enhancements for popular models and improved performance metrics.
Intel's LLM-Scaler project has rolled out a new update that engineers and developers working with AI models on Arc Pro B-Series graphics cards will find particularly appealing. The latest release brings support for Meta's Muse Glimmer 30B model with FP8 online quantization, marking a significant enhancement for those utilizing Intel's hardware for generative AI tasks.
The LLM-Scaler-vLLM beta 0.21.0-b3 release, unveiled on Monday, ensures same-day support for this new model, allowing users to immediately leverage its capabilities on the Arc Pro B70. Additionally, the update incorporates support for DFlash for Muse-Glimmer-30B and Qwen3.6-27B, offering users a broader range of options when deploying AI applications.
Performance enhancements are also on the docket, with improved time-to-first-token metrics for the Gemma-4-31B and Gemma-4-26B-A4B-it models. These tweaks are expected to facilitate smoother and more efficient AI operations, addressing some of the latency issues that users might have encountered in previous versions.
In parallel, the LLM-Scaler-Omni beta 0.2.0-b1 release introduces a revamped Docker container that upgrades to the ComfyUI 0.31 XPU stack. This update expands support for local video generation with MiniMax H3 and includes Wan Animate 2 for the Arc Pro B70 and B60 models. Furthermore, users will benefit from the expanded optimized model coverage, now featuring Wan 2.2 14B T2V Turbo, LTX-2, Z-Image / Lumina, and Krea2, alongside managed GGUF Q4_1 support and enhanced ComfyUI-GGUF-XPU integration.
These updates are part of Intel's ongoing efforts to streamline AI workloads and optimize their hardware for intensive computational tasks. Engineers working with Intel's Arc Pro series will find these improvements crucial for maintaining competitive performance in AI-driven environments.
The Linux Camp teaches these topics as hands-on labs on real virtual machines, verified as you type.