Furiosa SDK 2026.4 is now available, delivering up to 138% more tokens per watt across a wide range of models, extended prefix caching beyond NPU memory, support for vision-language models, and simplified model loading via Furiosa Executable Bundles (FXB).
Furiosa SDK 2026.4: Enhanced MoE performance, hierarchical caching, expanded model coverage
calendar_today
October 16, 2026
domain
furiosa