Local LLMs
Local LLMs are language models that run directly on your own hardware — a laptop, a PC with a GPU or a home server — with no cloud, no subscriptions and no data leaving your machine. Ollama, llama.cpp and LM Studio, plus quantized GGUF builds of Llama, Qwen and Gemma, make it possible even on a decade-old graphics card. Here is all our coverage of running AI locally: guides, hardware tests, quantization and privacy.
NEWHacker News AIOpen-source Hob brings Claude Cowork-style workflows to your laptop, no cloud required
RunPod BlogGenmo's Mochi 1: An Open 10B-Parameter Video Model Under Apache 2.0
Hacker News AIAI Agents Decompiled a First-Person Shooter, Burning 500 Billion Tokens
Habr AIReplacing LLM weights with a recipe: how 16 GB of matrices are being folded into formulas
Hugging Face Daily PapersThinkBit shrinks Qwen3-14B to 3.74 GB and still scores 94.8 on MATH-500
Habr AIRussia Already Has 18 AI Code Agents, but 'Russian' Means Three Different Things
Jiqizhixin (机器之心)Matrix Game 4.0 Hits 60 FPS on a Single Gaming GPU and Beats Nvidia's Rival Model
Jiqizhixin (机器之心)Stanford and Together AI Teach LLMs to Write Text Starting From an Outline: Generation Up to 31x Faster
ThinkBit shrinks Qwen3-14B to 3.74 GB and still scores 94.8 on MATH-500
ThinkBit's 2-bit scalar quantization preserves reasoning: Qwen3-14B shrinks from 29.5 GB to 3.74 GB and scores 94.8 on MATH-500.
Russia Already Has 18 AI Code Agents, but 'Russian' Means Three Different Things
RuBench's census counted 18 Russian AI code agents (10 for IDE/CLI and 8 builders): from GigaCode, SourceCraft, and Kodik to T-Code, Koda, and 1C:Naparnike. We…
Matrix Game 4.0 Hits 60 FPS on a Single Gaming GPU and Beats Nvidia's Rival Model
China's interactive world model Matrix Game 4.0 generates a playable image in real time at 60 FPS on a single consumer GPU.
PACE compresses linear LLM state to 4 bits and saves a third of the memory
Moonshot releases open Kimi K2 Thinking: 300 tool calls in a row
Moonshot open-sources Kimi K2.6: 300 sub-agents and 13 hours of autonomous coding
Moonshot releases Kimi K3 with 2.8 trillion parameters and a 1M-token context window
IF-MAP Finds Which Agents to Improve: Twice the Accuracy Gain
IF-MAP ranks agents in an LLM system by improvement potential: +3.81 pp from prompt optimization vs 1.85 with uniform selection, and up to 65% token savings.
A Frozen 16 MB Vocabulary Matches a Trained One in Small LLMs
Researchers replaced 168M embedding parameters in a Gemma-like model with a frozen 16 MB table and did not lose to the trainable version on three of four benchmarks.
NVIDIA releases Nemotron 3.5 Super: 26 AA points and 630 tokens/s
Developer Built Lumi with Claude Code: a Free AI-Powered Spotlight for Mac and Windows
Tech Aquarium on Two GTX 1660 Supers: How I Ran Qwen 2.5 and Built It Into INFERNUM
Alibaba's Qwen-Image-2.1: 7 Billion Parameters, 2K Output, and It Runs on a 16 GB MacBook Air
Fitbit Edge: Between Pixel Watch 5 and Fitbit Air — Google's New $180 Tracker
Google has launched the Fitbit Edge, an activity tracker slotting between the affordable Fitbit Air and the premium Pixel Watch 5. The device features on-device…
Drone Strikes on Yandex Infrastructure Cut Spare Computing Capacity in Russia
IT market analysts report a critical drop in spare computing capacity in Russia after a series of drone strikes on Yandex infrastructure.