Everything the lab has published, with the primary artifact attached: arXiv for papers, Hugging Face for weights and benchmarks, and the blog for the systems work behind them.
Filter by type or research area. Every entry links to its primary artifact.
Type
Research area
9 of 18 entries
2026 · Open model · Apache 2.0 (35B-A3B), research only (27B)
Action models at 27B dense and 35B-A3B that click, code, and call tools across desktop, web, and mobile, in BF16, FP8, NVFP4, and GGUF.
2026 · Open model · Apache 2.0
Multimodal encoders and retrievers for long-context document understanding, released at 260M and 800M parameters with dense and late interaction retriever variants.
2026 · Engineering note
How the infrastructure team traced and cut the startup time of the agent runtime, from model loading to warmup.
2026 · Engineering note
Finding and reclaiming GPU memory that stays allocated after a job ends, and the tooling built to catch it in production.
2026 · Dataset and benchmark · Apache 2.0
Screenshots paired with an instruction and start and end bounding boxes, so models can be trained and measured on drag and drop, an action most grounding datasets leave out.
2026 · Open model · Apache 2.0
Open-weight action models at 0.8B, 4B, 9B, and 35B-A3B, with quantized checkpoints for local and single-GPU deployment.
2026 · Open model · NVIDIA Open Model License
A computer-use model trained on NVIDIA's Nemotron 3 Nano Omni backbone, part of H's work on efficient open backbones.
2026 · Open model · Apache 2.0
The third generation of Holo action models, trained for grounding and multi-step control on web and desktop interfaces.
2026 · Engineering note
The cluster orchestration behind online reinforcement learning runs, where training and rollout environments have to scale together.