AI Briefing#local-llm
The latest AI and developer news about #local-llm, with the original source and a short summary.
The latest AI and developer news about #local-llm, with the original source and a short summary.
It's an MCP-based plugin that delegates heavy code generation tasks to a local LLM to reduce token consumption in Claude Code and Codex CLI.
A tool has been released that removes duplicate data within LLM context to increase effective token capacity.
TextGen has switched to an installation-free desktop app for Windows, Linux, and macOS.
A local 35B model handled about half of daily tasks faster than the cloud.
The number of GGUF-format model uploads on Hugging Face has roughly doubled over the past two months.
Triggered by GitHub Copilot's billing shift, local LLM hardware is proposed as a solution for AI self-sufficiency.
BeeLlama.cpp, a new fork that maximizes the performance of llama.cpp, has been released.
cdo, an open-source tool that lets you use local Ollama models while keeping the Claude Desktop UX, has been released.
AMD has unveiled the MI350P PCIe AI GPU, which can be plugged into existing servers.
This provides a comparison of open source translation tools that utilize the Gemma 3-based TranslateGemma model.
Thoth, a local-first AI assistant, has announced the v3.20.0 update, which includes Linux support and MiniMax integration.
AIMEAT, an open protocol that lets humans, AI agents, and local LLMs share apps, knowledge, and functions, has been unveiled.
It's a RAG server that uses local embeddings and MCP to identify semantic relationships between notes and automatically inject context into Claude.
It is an open-source library for RAG that indexes over 200 million web pages, enabling local web search without any API calls.
claudely, an open-source tool that lets you connect local LLMs while keeping Claude Code's ecosystem, has been released.
Running Mistral Medium 3.5 128B Q3_K_M locally on 3x3090 (72GB VRAM) measured 9.67t/s.
A Strix Halo 128GB system ran Mistral Medium 128B at 1.57 tokens per second.
A reproducible setup was shared for running Qwen 3.6 35B MoE on an RTX 3090 with 262K context.
Ran Qwen3.6-35B-A3B at 23 t/s on a 5-year-old laptop with 6GB VRAM.
The function calling harness has narrowed the frontier-local gap in backend generation.
AutoBe has released a monthly benchmark comparing backend generation by GLM, Qwen, and DeepSeek.
OpenJet v0.4 has been released, strengthening its llama.cpp-based local coding agent capabilities.
Qwen3.6-27B achieved 95.7% on SimpleQA in a single RTX 3090 local environment.
Ran Qwen 35B-A3B at 8.9 TPS on a 16GB M3 MacBook Air using `mmap`.