[Dev Weekly #116] GLM-5.2 Dethrones the Giants | Claude’s Becoming Difficult | Rust’s Hidden Memory Trap

by Gabriel Quaresma on June 19, 2026 5 min
Table of Contents

HELLO EVERYONE!!! It’s June 19th, 2026, and you are reading the 116th edition of Codeminer42’s tech news report. Let’s check out what the tech world showed us this week!

Introducing the MDN MCP server

Mozilla released an experimental MCP server that connects MDN’s documentation and browser compatibility data directly to AI tools and IDEs. The testing shows it dramatically improves accuracy on browser support queries and cuts response times in half, letting you rely on current web platform information instead of training data cutoffs.

RFC 10008: The HTTP QUERY Method — by J. Reschke, J.M. Snell, M. Bishop

The IETF standardized a new HTTP QUERY method that bridges GET and POST for safe, idempotent queries with request bodies. Unlike GET, QUERY can handle large payloads without URI encoding overhead; unlike POST, it signals safety and idempotency to enable caching and automatic retries. If you’re building APIs or working with HTTP semantics, understanding this addition helps you choose the right method for your query patterns.

SpaceX locks in $60 billion Cursor deal to close gap with rivals in AI coding race

SpaceX has secured a $60 billion deal with Cursor to accelerate its progress in the AI coding sector and narrow the gap with competitors. This major agreement highlights rapid growth and features support from high-profile backers. The move marks a significant strategic shift as the aerospace giant expands into artificial intelligence development. The deal aims to leverage Cursor’s capabilities to enhance SpaceX’s internal coding and engineering efficiency. This development underscores the increasing intersection between space exploration and advanced AI technologies.

GLM-5.2 is probably the most powerful text-only open weights LLM — by Simon Willison

Z.ai just released GLM-5.2, a 753B parameter open weights model that’s now topping the Artificial Analysis benchmarks and ranking second on the Code Arena WebDev leaderboard. It features a million-token context window and costs significantly less than GPT or Claude, though it does burn through output tokens at a higher rate. You’ll see how this stacks up against proprietary models and get a sense of its actual capabilities from hands-on SVG generation examples.

Shopify Joins the Ruby Alliance

Shopify has joined the Ruby Alliance, bringing years of experience operating one of the world’s largest Ruby applications. Their investment signals the ecosystem’s maturity and gives the Alliance valuable perspective on scaling Ruby at global proportions.

Scaling Rails at Aura Frames: Splitting to 8 Primary DBs and Reaching #1 in the App Store – by Andrew Atkinson

Aura Frames scaled its Ruby on Rails platform to #1 App Store ranking by splitting a single PostgreSQL database into eight primaries, using Rails’ disable_joins to handle cross-database queries without SQL joins. This architecture, combined with vertical scaling and custom caching strategies, allowed the team to handle 41 million API requests per hour during Christmas 2025 without reliability issues. The migration maintained a single codebase while successfully managing massive traffic spikes through application-level sharding.

Your Rust Service Isn’t Leaking — It Could Be the Allocator — by Pranitha

When a Rust service’s memory stayed pinned at container limits despite no actual leaks, the culprit was glibc’s memory allocator fragmentation pattern. Learn how thread arenas and unconsolidated chunks can trap memory even after tasks complete, and why switching to jemalloc with background threads solved the problem for bursty, concurrent workloads.

Gemma 4 12B Coder Fable5 Composer2.5 v1 GGUF — by yuxinlu1

A quantized version of Google’s Gemma 4 12B model optimized for code generation and composition tasks. This GGUF format makes it practical to run locally on consumer hardware while maintaining strong coding capabilities, giving you a capable alternative to cloud-based coding assistants.

zlib-rs in Firefox — by Folkert de Vries

Firefox 151 now uses zlib-rs for gzip compression, delivering significant performance gains (up to 32x faster decompression on some workloads). Learn how Mozilla’s integration surfaced an Intel Raptor Lake CPU bug, the creative workaround the team implemented, and why the two-year journey was worth the effort.

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

Anthropic was forced to disable Fable 5 and Mythos 5 globally after a US government export control directive citing a potential jailbreak technique. The company pushes back, arguing the vulnerability is narrow, exists in other models, and doesn’t warrant pulling a commercially deployed system, and they’re working to restore access quickly.

SVGs and PDFs can both be interactive

Both SVG and PDF specs hide rarely-used interactive capabilities that let you embed JavaScript and create self-contained apps without a web server. You’ll discover how to leverage these features, why most tools skip them entirely, and concrete examples of draggable shapes working in both formats.

Why Is Claude Turning Into An Asshole? — by Bram Cohen

Bram Cohen breaks down why Claude’s recent versions (especially Fable) have become increasingly argumentative and difficult to work with, tracing the shift to over-aggressive alignment guardrails, poor semantic nitpicking, and training priorities favoring coding over conversational ability. If you rely on Claude for anything beyond pure coding tasks, you’ll find concrete examples of how the model’s behavior has degraded and what might actually fix it.

Languages, Tools & Framework releases

Open Knowledge Format v0.1 — by Sam McVeety, Amir Hormati

Google introduces the Open Knowledge Format, an open specification for representing metadata and context that AI agents need, built on simple markdown files with YAML frontmatter. If you’re building agents or dealing with fragmented knowledge across your organization, OKF offers a vendor-neutral way to make that knowledge portable and shareable without SDKs or proprietary lock-in.

hurried — by yankouskia

A lightweight JavaScript library that wraps Node.js worker_threads with a friendlier API for parallel code execution. You get typed RPC, event buses across worker boundaries, pools with backpressure, streaming results, and retry logic all behind a minimal surface area that scales from single tasks to multi-CPU pools without rewriting code.

—

And that’s all for this week! Wish you all a great weekend and happy coding!

We want to work with you. Check out our Services page!