Khronos royalty-free open standards for 3D graphics, Virtual and Augmented Reality, Parallel Computing, Machine Learning, and Vision Processing

Apex Compute is building high-performance, low-power AI chips for edge devices and developing GPU architecture. Their AI accelerator is designed to seamlessly run modern transformer models directly on devices like robots, drones, smart cameras, wearables, and autonomous vehicles. Apex Compute's GPU architecture is being developed with the goal of Vulkan conformance and already runs graphics workloads at the FPGA prototyping stage. They are creating more efficient alternatives to traditional computing architectures, with a focus on low latency, strong real-world performance, practical deployment, and architectural efficiency.

GRENETA is a South Korean deep-tech company developing 3D spatial data infrastructure for Physical AI, robotics, and industrial simulation. Their technology transforms real-world environments into high-fidelity, lightweight 3D assets and scalable spatial datasets, bridging the gap between physical-world capture and AI-driven simulation. At the core of GRENETA's technology is their proprietary GRENETA Optimizer, which addresses one of the major bottlenecks in large-scale 3D applications: excessive data size, transmission costs, and GPU memory consumption. In selected tests, their technology has achieved up to 99.6% data size reduction while preserving high visual fidelity.

Beyond compression, GRENETA is developing an integrated workflow encompassing 3D reconstruction, optimization, format conversion, and spatial data structuring. Their long-term objective is to enable Simulation-Ready and Robot-Ready 3D datasets, incorporating object-level semantics, spatial relationships, and physics-relevant attributes for Physical AI applications. GRENETA's vision is to make real-world 3D environments computationally efficient, accessible, and usable as foundational spatial data for the next generation of Physical AI.

The Khronos OpenXR Working Group has released a new set of extensions that establish an open, cross-platform API for spatial windowing systems. The extensions introduce "spatial containers" to OpenXR, which are system-managed spaces in which applications can display XR content, similar to how traditional applications draw 2D content within windows on a desktop. Spatial containers give developers a standardized way to create applications that can move between immersive experiences and multitasking with the real world or other apps across compatible XR platforms.

MESA In development the past year has been the Mesa Gallium3D API implemented atop NVIDIA's CUDA driver API. Initially named "CLUDA" and now going by "NOCL", this Mesa code paired with Rusticl is working well for OpenCL atop the NVIDIA proprietary driver as a new compute option.

KosmicKrisp, LunarG’s Mesa Vulkan driver for Apple silicon, has passed the Vulkan 1.4 Conformance Test Suite (CTS) and is now a Khronos-conformant Vulkan 1.4 implementation. The Vulkan 1.4-conformant version of KosmicKrisp is available for download via the Vulkan SDK.

KosmicKrisp was already a Khronos-conformant Vulkan 1.3 product. Vulkan 1.4 conformance is the next step in advancing support for Vulkan on macOS. In addition to reaching 1.4 conformance, substantial support for non-core extensions and features has been added, including Tesselation, Robustness2, and Multi-draw. KosmicKrisp is implemented as a Mesa driver, shipped in the LunarG Vulkan SDK, and exercised in LunarG’s CI. Performance of KosmicKrisp has tripled since it achieved Vulkan 1.3 conformance, and optimization work continues to be a high priority.

The OpenCL Working Group has published the OpenCL 3.1.2 specifications, finalizing the cl_khr_command_buffer extension alongside bug fixes and clarifications. Command buffers let applications record a sequence of commands once and replay it repeatedly, reducing CPU overhead and giving drivers opportunities to optimize across the entire sequence. This benefits workloads with repeated execution patterns, including AI inference. The specifications are available in the OpenCL Registry.