A mysterious model is going viral among developers

Welcome back. A mysterious new model is growing on developers. While its creators remain in the shadows, devs are flocking to it for its uncensored output and massive rate limits. Experts suspect it hails from a Chinese AI lab, but it’s still a secret. Read the details below. 

Also: ChatGPT can now send iMessages, a senior engineer’s complete development workflow (tutorial), and a fun weekend project to play with virtual light.

A mysterious model is going viral among developers: OpenCode just unveiled Ox Alpha, a stealth AI packing a massive 1M token context and nearly unlimited rate limits. Technical fingerprints suggest it might be a secret multimodal variant from a major Chinese lab, but the makers of the model still remain in the shadows. In an early testing attempt, a developer reported an 80% success rate on coding tasks. OpenCode is inviting developers to try it out.

Anthropic ships Claude Academy to train dev teams: The AI lab just launched a self-paced learning hub called Claude Academy. You can learn Claude, APIs, prompting, and agent skills using material adapted from Anthropic’s own internal training. You can start learning Claude Code here. Anthropic didn’t stop there. They also pushed Computer Use, the Browser Tool, and the Skills and Files APIs to general availability, giving engineering teams the production-ready building blocks they need to actually ship AI agents. 

ChatGPT can now send iMessages on your behalf: OpenAI just dropped an Apple Messages plugin for the ChatGPT Mac app. It works inside Codex and ChatGPT Work and can search, summarize, draft, and send iMessage, SMS, and RCS threads. It doesn't work in regular chats. Like any new AI feature, it’s sparking privacy discussions on the internet. So far, OpenAI says everything runs locally and they aren't indexing your messages, but the technical specifics are still pretty vague. 

Growing Postgres databases eventually hit a wall: queries slow, dashboards lag, and adding another database starts to feel inevitable.

Tiger Data keeps Postgres fast as your data grows with time-series primitives, columnar storage, and automatic partitioning. Same SQL and tooling. No pipelines or second system.

Just Postgres, built to scale with your workload.

Start building for free today. No credit card needed.

The undeniable truth. APIs are how your product connects to the world, but a messy underlying system will always bleed through a clean interface. GitHub engineer Sean Goedecke's API design post is making the rounds again and with good reason. Your API is only as good as the system beneath it. His rules worth keeping in mind while building an API:

Why this matters. Building the simplest API possible is the goal, but you also need to bake in long-term flexibility. Great API design is really just finding the sweet spot between those two. Nail that balance, and your API becomes a major reason why users stick around. 

PS: If you’re ready to start building, a solid API design roadmap is a good place to start.

The deploy question that comes up every sprint. The incident runbook saved somewhere nobody can find.

Multiply that across every team you lead, and the same five questions get answered in Slack — never written down anywhere. 

Scribe turns how your teams actually work into step-by-step guides, automatically, as they work.

See how it works

Senior engineer’s complete development system: You’ll learn how to build and ship full-stack SaaS applications using AI models, agentic development environments, custom agent skills, and automated code reviews.

VS Code inside the terminal: This tool brings a VS Code-compatible editor straight into the terminal. It works over SSH, syncs with your terminal theme, and gives you a familiar coding UI without leaving the command line.

Build agents that can go the distance (by Google): If agents are going to run for hours or days, they need to survive failures without starting from scratch. This tutorial breaks down five design patterns for long-horizon agent harnesses, covering everything from checkpointing and resuming work to pausing for human approval. 

Local models worth using normally need 30GB+ of RAM. A 4-bit Qwen3.8-27B needs 17GB+. Unsloth's new Dynamic 3.0 quants shrink it to ~6.5GB at 1-bit, retaining 77% top-1 token accuracy by Unsloth's own measure. Fine for chat and quick questions; Unsloth says don't use it for tool calling or agentic work.

Got 16GB? Use UD-Q2_K_XL (~9.8GB), which Unsloth recommends as the real floor.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

A dev’s Mac battery recently tanked from 100% to 10% in under three hours. The culprit wasn’t the terminal, and the fix is worth a look if your battery has been struggling too. 

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team