Local coding-agent roundup says Gemma 4 26B-A4B beats dense 24B models in refactoring tests
craigsdennis · x · 2026-07-28
A shared roundup collects several useful resources on local model coding agents, with a strong focus on Gemma 4.
Key points from the linked material:
- A refactoring eval found 14–24B dense local models at 0/10, while Gemma 4 26B-A4B jumped to 9/10.
- The model is a 26B MoE with only 4B active per token, which makes it much faster in practice.
- Another guide explains how to wire local models into coding harnesses such as Claude Code.
- In tool-reasoning tests, gemma4:e2b reportedly scored 0/5, while Qwen 35B-A3B solved 4–5/5.
- There is also a walkthrough for configuring pidotdev with Gemma 4, plus a debate about whether parameter count or active parameters matter more.
The thread reads like a compact map of the current local-coding-agent stack: evals, harnesses, model tradeoffs, and setup paths.
More from coding & agent
- Loop engineering uses a frontier model to build skills, then loops GLM 5.2 calls until it works — steipete · 2026-07-28
- Mnemos is getting a rebuilt desktop app, new MCP tools, and a redesigned platform — RileyRalmuto · 2026-07-28
- Perplexity Computer and Comet Assistant built 7 GA4 reports in 90 minutes — gaganghotra_ · 2026-07-28
- Codex hits a session limit after one shell command in a coding workflow — AIandDesign · 2026-07-28
- Distributed Systems open-sources DSCO, a C-written agentic LLM CLI with MCP and swarms — arthurcolle · 2026-07-28
- Salesforce’s StateAct lifts Opus 4.8 on OSWorld 2.0 with state-first agents — Salesforce · 2026-07-28