NEWS

Claude Code request marking: hidden unicode apostrophe and date-format markers flag timezone, domain and lab keywords. What was found and what to do about it.

Anthropic releases Claude Sonnet 5 with improved agentic capabilities, better tool use, and an introductory pricing deal. Here's what developers need to know.

Ornith-1.0 open-source coding models explained: what "self-improving" really means, the four model sizes, hardware needs, HN reactions and whether to try it.

Outer Shell proposes a graphical desktop for remote servers over SSH, with apps served on Unix domain sockets. How it works, the HN debate and who it is for.

PostgreSQL 19 beta features: SQL/PGQ graph queries, SQL:2011 temporal tables and REPACK CONCURRENTLY, plus what HN thinks and when to expect the release.

ZLUDA 6 runs unmodified CUDA apps on AMD GPUs and adds PhysX and Blender textures. How it compares to ROCm and HIP, and why hobby status matters.

A developer fed 266MB of DICOM MRI data to Claude Code for a second opinion on a shoulder diagnosis. It disagreed with the doctor; HN radiologists weighed in.

GLM 5.2 vs Claude Code on Semgrep IDOR benchmarks: the open-weight model scored higher at about one-sixth the cost per finding. Setup, results and HN reactions.

Google Workspace CLI creator Justin Poehnelt saw his tool go viral and hit #1 on Hacker News, then was fired two days after Google announced its own version.

Filippo Valsorda argues LLMs have made vulnerability reports less special. His case against old coordinated disclosure norms, plus the HN pushback.

Armin Ronacher's The Coming Loop on agent loops versus harness loops: where loop-driven development works, where it fails, and what the HN thread adds.

Run GLM-5.2 locally with Unsloth dynamic quantization: hardware math for a 256GB Mac or 24GB GPU, quantization tradeoffs and lessons from the HN thread.

Memcached vs Redis for caching: the architectural case that memcached's constraints are a feature, the HN pushback, and when you need either one.

TikZ Editor is an open-source WYSIWYG editor for LaTeX TikZ figures built almost entirely by Codex. How it works and what HN made of agent-built software.

Unlimited OCR is Baidu's open-source vision-language model that parses 100+ page documents in one pass without memory blowup. How it works and when to use it.

VibeThinker-3B is a 3B parameter reasoning model reporting 94.3% on AIME 2026. What the paper claims, how to run it locally, and what it trades away.

Switzerland's fully open foundation model promises transparent training data and EU compliance. The HN crowd has questions about actual performance.

A developer discovered that Claude Code's thinking output is summarized, not the raw reasoning. Here's what Anthropic's docs actually say - and why it matters.

A Codex CLI SQLite logging bug wrote TRACE logs that could wear out SSDs. What caused it, the merged fixes, and what to check if you run Codex heavily.

Deno Desktop in Deno 2.9 compiles TypeScript projects into native apps with a system WebView or bundled Chromium. How it compares to Electron and Tauri.