Local LLM
Open-source models to run on your own machine: Llama, Mistral, Qwen, Gemma, Phi, DeepSeek… Quantization, fine-tuning, hardware optimizations, benchmarks and inference tools (llama.cpp, ollama, vLLM).
Local LLM Kimi 3 aims to challenge Opus 4.8 with a Chinese open giant
Moonshot is reportedly preparing Kimi 3, a Chinese open model with 2 to 3 trillion parameters, designed to rival Claude Opus 4.8.
Read the article
Local LLM Hugging Face wants to speed up LLMs with diffusion
Hugging Face is highlighting diffusion language models with Nemotron-Labs, a promising path to dramatically speed up text generation.
Read the article
Local LLM Hugging Face speeds up LLMs with Nemotron-Labs
Hugging Face is highlighting Nemotron-Labs’ diffusion language models, a promising path to significantly speed up text generation.
Read the article
Local LLM Hugging Face wants near-instant LLMs through diffusion
Hugging Face is highlighting Nemotron-Labs’ diffusion language models to greatly speed up text generation and challenge autoregressive LLMs.
Read the article
Local LLM Open-source LLMs locally: which ones run on a PC in 2026?
2026 editorial comparison of open-source LLMs that are genuinely usable locally on a standard PC: RAM, VRAM, speed, quality, privacy, and trade-offs.
Read the article
Local LLM DiffusionGemma: Google promises text 4x faster
Google DeepMind introduces DiffusionGemma, an approach that promises text generation up to 4x faster than traditional LLMs.
Read the article
Local LLM GLM-5.2 aims to make open source matter for long tasks
Hugging Face introduces GLM-5.2, a model designed for long tasks and persistent agents, with an open-source positioning worth watching.
Read the article
Local LLM Hugging Face simplifies one-click vLLM deployment
Hugging Face now makes it possible to launch a vLLM server with a single command via HF Jobs, a concrete step toward industrializing open source LLMs.
Read the article
Local LLM GLM-5.2 aims to rival Anthropic in cybersecurity
Z.ai unveils GLM-5.2, an open-weight model presented as capable of rivaling Mythos on certain cybersecurity tasks.
Read the article