Running Google Gemma 4 26B-A4B Locally with LM Studio 0.4.0 Headless CLI
What LM Studio 0.4.0 Adds for Local AI
LM Studio 0.4.0 fundamentally changes the architecture by extracting the core inference engine into llmster, a standalone server. This enables running LM Studio entirely from the command line using the new lms CLI, eliminating the need for the GUI. The update makes it usable on headless servers, in CI/CD pipelines, SSH sessions, or for terminal-focused developers.
Key Features in 0.4.0
- llmster daemon: A background service that manages model loading and inference without the desktop app
- lms CLI: Full command-line interface for downloading, loading, chatting, and serving models
- Parallel request processing: Continuous batching instead of sequential queuing, allowing multiple requests to the same model to run concurrently
- Stateful REST API: A new /v1/chat endpoint that maintains conversation history across requests
- MCP integration: Local Model Context Protocol support with permission-key gating
Why Gemma 4 26B-A4B for Local Use
Google's Gemma 4 26B-A4B uses a mixture-of-experts architecture with 128 experts plus 1 shared expert, but only activates 8 experts (3.8B parameters) per token. This means it runs well on hardware that couldn't handle a dense 26B model. On a 14" MacBook Pro M4 Pro with 48GB unified memory, it fits comfortably and generates at 51 tokens/second.
The model scores 82.6% on MMLU Pro and 88.3% on AIME 2026, close to the dense 31B variant (85.2% and 89.2%) while running dramatically faster. It achieves an Elo score of ~1441, competing with models like Qwen 3.5 397B-A17B (~1450 Elo) that require 100-600B total parameters.
Key capabilities include 256K max context, vision support for analyzing screenshots and diagrams, native function/tool calling, and reasoning with configurable thinking modes.
Practical Setup
The article walks through installing the lms CLI and setting up Gemma 4 26B-A4B for local inference that can be used with Claude Code. The author notes significant slowdowns when used within Claude Code from their experience.
📖 Read the full source: HN AI Agents
👀 See Also
Clarc v1.0: Workflow OS for Claude Code with 63 Agents and 249 Skills
Clarc is a plugin layer for Claude Code that provides 63 specialized subagents, 249 domain skills, and 178 slash commands for development workflows. Installation is via npx with support for multiple editors including Cursor and OpenCode.
RalphTerm: ralph-style loop for Claude Code with cross-review sessions from different agents
RalphTerm is an open-source Rust CLI that runs a ralph-style outer loop around Claude Code: it takes a markdown plan, executes tasks in fresh interactive sessions, and runs cross-review with a different model (e.g., Codex) in separate fresh sessions, feeding issues back into new implementer sessions.
Banker Creates Credit Due Diligence Tool with 31 AI Prompts Using Only Claude
A banker with 17 years of MSME credit underwriting experience in India has open-sourced 31 AI prompts that transform a visiting card into a full credit due diligence report, compressing a 3-4 week process to 30 minutes. The tool was built entirely through conversation with Claude, without writing any code.
Cowork Chrome Extension Automates Personal Data Removal from Data Brokers
A Reddit user reports that using the Cowork Chrome extension with a Gmail connection automated filling forms, writing emails, and verifying removal requests to delete personal data from major data providers in just a few hours.