efficient

SANA World Model is a new AI model that uses hybrid linear attention for efficiency and speed.

SANA-WM is a 2.6B-parameter open-source world model that generates high-fidelity 720p minute-scale videos with precise camera control, achieving industrial-level quality while significantly reducing computational requirements.

The paper introduces δ-mem, a lightweight online memory mechanism that augments frozen LLMs with a compact associative memory state updated by delta-rule learning, achieving significant improvements on memory-heavy benchmarks without fine-tuning or context extension.

OpenAI released a 1.5B-parameter MoE model with only 50M active parameters that can filter private data from trillion-token datasets while maintaining 128k context length.

NVIDIA introduces Nemotron OCR v2, a fast multilingual OCR model built using synthetic data generation. The model achieves 34.7 pages/second on a single A100 GPU by using a unified FOTS-based architecture with feature reuse across detection, recognition, and relational components.