Latest
Latest from OpenModelsHQ
Editorial picks, practical research and recent records from the model catalogue.
Research and guides
Is DeepSeek safe to use? Local weights vs the app and API
Local DeepSeek weights keep your data on your machine. The app and API store it in China under DeepSeek's privacy policy. Restrictions and risks.
Which open models run on the NVIDIA DGX Spark
The DGX Spark has 128 GB of unified memory. See how that decides which open models fit, how clustering works, and when renting is the better choice.
Who owns DeepSeek, Moonshot AI, Zhipu (Z.ai) and MiniMax
High-Flyer owns DeepSeek, Alibaba took a reported 36% of Moonshot, and Zhipu and MiniMax trade in Hong Kong. Founders, funding and listings.
Run open models with Ollama
How to install Ollama on macOS, Linux and Windows, run open models, pick quantised tags, pull GGUF from Hugging Face and use the OpenAI-compatible API.
How to choose a GPU for local LLMs
Pick a GPU for local LLMs by memory first, bandwidth second and software support third. Consumer cards, the RTX PRO 6000, Macs and the DGX Spark.
Xiaomi MiMo explained: models, licences and where it fits
Xiaomi's MiMo models run from a 7B reasoning model to a 1.02 trillion parameter one, mostly under MIT. Release history, API-only launches and licences.
RTX PRO 6000 Blackwell for local AI
The RTX PRO 6000 Blackwell has 96 GB of GDDR7. Specs for each edition, what that memory makes possible, and how it compares with the RTX 5090.
GGUF and llama.cpp: run quantised open models
What the GGUF format is, how to run it with llama-cli and llama-server, how to read quant names like Q4_K_M, and how LM Studio and Ollama relate.
Run open models with MLX on a Mac
How to run open models on Apple silicon with MLX and mlx-lm: install, mlx_lm.generate, the local server, mlx-community repos, unified memory and mlx-vlm.
AMD ROCm and HIP for AI: what is supported
Which AMD GPUs ROCm officially supports, the PyTorch and vLLM versions AMD validates, and what HIPIFY does to CUDA source.
Huawei Ascend, CANN and other non-CUDA accelerators
What CANN is, how torch_npu and MindSpore reach Ascend NPUs, the versions vLLM Ascend pins, and where Tenstorrent, TPU and AWS Neuron stacks sit.
Intel oneAPI, SYCL and PyTorch XPU support
What SYCL and oneAPI are, which Intel GPUs PyTorch validates for the xpu device, and what Intel's notice says about the Intel Extension for PyTorch.
llama.cpp backends: Vulkan, CUDA, Metal, SYCL and more
The full llama.cpp backend list, which ones ship prebuilt binaries, and why Vulkan is the practical choice on consumer AMD and Intel GPUs.
Metal, MPS and MLX: the Apple silicon AI stack
How Metal, the PyTorch mps device and Apple's MLX framework fit together on a Mac, with the macOS and hardware requirements each one states.
ZLUDA and CUDA translation layers: where things stand
What ZLUDA translates, which AI workloads its developer reports working, and how funding changes affected its scope.
CUDA explained: why it became the default for AI
What CUDA is, how cuDNN, NCCL and TensorRT deepen the dependency, and what the CUDA Toolkit licence actually permits, from NVIDIA docs and EULA.
Triton and Mojo: portable kernels above CUDA
OpenAI Triton as a kernel language PyTorch already depends on, and what Modular claims for Mojo and MAX, with the platform support each one documents.
Open audio models: speech to text, speech and music
How open ASR models are measured by word error rate, what faster-whisper and whisper.cpp do differently, and where speech synthesis and music sit.