Optional Google Analytics and advertising are off until you choose. Read our privacy details.
Topic
3 news articles · 7 tutorials
In mid-June 2026 Z.ai shipped GLM-5.2, a 753-billion-parameter open-weight model that runs roughly even with GPT-5.5 on long-horizon coding. The card on my desk holds 6 GB. After months of running Gemma 3 4B vision on a GTX 1660, this is the real cost of self-hosting open models in VRAM, latency, and quality.
Heretic 1.3 lands reproducible runs, an integrated benchmark harness, and lower peak VRAM. The real shift: forks dressing up the same technique in jargon just got transparency-pressured into the open.
Ollama's v0.21.0 drops Hermes Agent and Copilot CLI integration, but the real win is smoother local inference on Macs and cleaner config management, a quiet upgrade cycle that matters more than the flashy features.
Cline is a VS Code extension that turns any LLM endpoint into a coding agent. I wired it to a local Ollama Qwen 2.5 7B and ran a week of work without a single API call.
GPT4All is the simplest local-LLM desktop app I recommend to first-time users. Three platforms, one click, models running offline. This is the install plus a real first-day workflow.
Jan is the open-source desktop chat app that runs entirely on your machine. I tried it for a week as a ChatGPT replacement. This is the install plus the workflow that worked.
LM Studio is the friendliest local-LLM front-end I have used. Built-in model browser, one-click downloads, OpenAI-compatible API. This is the install plus the workflow I use on the MacBook.
I ran Gemma 4 9B locally on a 32GB Linux box for a week of normal use. This is the Ollama install, the model variants worth pulling, and the one config tweak the docs do not flag.
Qwen 2.5 7B is my preferred local coding model. The pull is 4.4GB, which on Indian fibre means a 12-20 minute commitment. This is the install plus the bandwidth-saving tricks I use.
Phi-3 Mini on a Pi 5 is the cheapest serious LLM setup you can build. I ran one for two weeks. This is the install, the real throughput numbers, and the use cases that work.