10 articles
NVIDIA’s new open Nemotron 3.5 Super VL and Bolt models shift the agent race toward efficient multimodal inference, reproducible recipes, and deployment cost.
Liquid AI's LFM2.5-24B-A2B release makes local agent design a memory, latency, and control problem rather than a cloud-model shopping exercise.
Reports that NVIDIA is considering a deeper investment or acquisition of Reflection AI reveal how open-weight model competition is becoming an infrastructure and distribution contest.
LightOnOCR-3 adds grounding, chart extraction, and image descriptions to lightweight OCR, changing what teams can safely do with documents on local hardware.
Liquid AI’s open d1-3B and d1-omni-600M models answer structured questions in one forward pass, bringing multimodal decision systems to edge hardware.
H Company’s Holo4 models combine GUI actions, code, MCP, and APIs; the release makes interface switching and reproducible trajectories central to agent evaluation.
Black Forest Labs’ FLUX.3 Action release points from image generation toward action-conditioned visual models for robotics and interactive systems.
NVIDIA's Nemotron work suggests open models are becoming a strategic tool for enterprises and governments that want control as much as capability.
Google's new DiffusionGemma release is less about beating every benchmark and more about proving that speed, editability, and inference efficiency can justify a different model architecture.
NVIDIA Ising open AI models target quantum calibration and error correction, positioning AI as the control layer for practical quantum computing.