What changed in AI today

• Granite-Switch-4.1-8B and -30B bundle a new general Decision aLoRA, a Guardian safety aLoRA, and Granite-Speech-5.0-Turbo-CTC for transcription.
• All components fit in roughly the same GPU VRAM as the base model alone, since adapters share its KV cache.
• Adapters are Activated LoRAs, so a base model generation's KV cache can be reused for later adapter decisions.
• Users can compose their own Granite Switch models from Granite Libraries adapters; an Adapter SDK for Mellea is planned.

Why it matters for marketers: Teams running chat, moderation and transcription can consolidate onto one hosted endpoint and fewer GPUs, and can build custom adapter mixes.

• One model handles text-to-image and multi-reference image-to-image, plus new bounding-box layout control via JSON regions in the prompt.
• Distilled checkpoint reportedly makes a 1024 x 1024 image in about 2 seconds on an H100.
• NVFP4 and FP8 weights load with from_single_file for smaller GPUs; LoRA training scripts cover text-to-image and editing.
• Ships as a Modular Diffusers pipeline, and the base checkpoint is the one recommended for fine-tuning.

Why it matters for marketers: Content teams can prototype layout-controlled images and train custom LoRAs on open weights, with reported speed figures to plan GPU budgets.

• Mixture-of-experts model with about 816B total parameters, roughly 42B active per token.
• Accepts text, images, audio and video; context window reaches up to one million tokens.
• Meta says it was trained for agentic coding, including multi-file refactors and debugging.
• Two checkpoints are listed: a BF16 version and an NVFP4 version aimed at NVIDIA Blackwell GPUs.

Why it matters for marketers: Content and developer teams can self-host a long-context multimodal model and test it on screenshots, voice notes and screen recordings alongside code.

• Ai2's AI Infrastructure team swapped a priority-based scheduler for GPU time budgets, hierarchical fair-share allocation, and a time-slicing contract.
• Clusters of 88 to 1024 GPUs serve about 150 internal researchers, with outstanding requests reportedly 2-3x available GPUs.
• Workloads must declare a minimum runtime; unbudgeted, zero-runtime jobs are free but always preemptible.
• Automated draining of unhealthy hosts reportedly cut repairs needing a human-in-the-loop by 74%.

Why it matters for marketers: Teams sharing GPU clusters can see how budget-based allocation and minimum runtime contracts reduce idle capacity and on-call toil.