Qwen3.5-0.8B run in pure Assembly hits 6.51 tokens/sec

STAT reports newly surfaced documents add scrutiny to Medicare's AI-driven prior-authorization experiment, though the article is exclusive to STAT+ subscribers and the specifics are behind the paywall.

Hugging Face's blog post walks through LoRA-based fine-tuning of 3D generation models, per its URL slug 3d-lora-finetuning. No benchmark numbers or model names were included in the available snippet.

Style Advisor on JCrew.com recommends complete outfits from the brand's own catalog photo shoots, live now for women's and men's styles in the U.S. Built in Claude Code, it uses Fable 5.1 for planning, Opus 5.5 for the prototype, Sonnet 5.5 as the stylist, and Haiku 5.5 for fast tasks.

Nebius Group NV dropped about 10% in early Thursday trading after the AI computing provider issued a disappointing sales outlook for the year.