New Grok model upgrades coding without raising the price

Welcome back. We've got a packed release day today. Grok 4.7 just posted major benchmark gains at zero extra cost, Xiaomi shipped two new dev-focused models, and you can now run Hermes directly on your existing Claude Pro or Max plan. 

Also: A tutorial on how to ship 2500 PRs a month, Amazon bans Meta Muse from shopping on their website, and how OpenAI’s Astra could outperform Anthropic’s Fable.

New Grok model gets better at coding without a price hike: Elon Musk’s lab just dropped Grok 4.7, its flagship model for coding, agents, and knowledge work. Built on a larger base with longer reinforcement training, it self-verifies outputs, handles longer context, and runs agentic tasks through Grok Build. Compared to version 4.6, xAI's own benchmarks saw a massive jump: Terminal-Bench nearly doubled to 38.0%, while legal hit 19.6%. Even better, pricing holds steady at $2 per million input tokens and $6 per million output tokens, matching the previous version. Try it here.

The viral Hermes agent can now run on your Claude plan: Hermes creator Teknium just shipped an experimental plugin called Claude Subscription DirectSDK. It lets Hermes tap into your existing Claude Pro or Max subscription so you can dodge pay-per-use API bills. The plugin runs the actual Claude app in the background, meaning you don't need an API key. Just a heads up, Teknium says it burns through your allowance roughly 1.7x faster than using Claude Code directly. Try it here.

Xiaomi hands devs an open model built for agent stacks: The Chinese electronics giant just released MiMo-V2.6-Pro and its lighter sibling, Flash. Both models are MIT-licensed and available on Hugging Face. They handle complex software tasks with coding, tool use, multimodal input, and a massive 1M-token context window.  Flash delivers much of Pro’s agent performance at a fraction of the cost, making it a compelling option for teams running thousands of agent calls. Find the API here.

Not because you don't care. Because after 6 hours of building, typing paragraphs about what you built feels like punishment.

So the PR description stays vague. The README gets "TODO." The architecture decision lives in your head until you leave the company.

Wispr Flow: syntax-aware voice dictation that works in every app. Speak your PR descriptions, commit messages, and docs. camelCase, snake_case, and acronyms stay intact.

Engineers at OpenAI and Vercel use it daily. Available on Mac, Windows, iPhone, and Android. The best docs are the ones that actually get written.

Try free

After the promotion. You land the senior role; now you have to prove you deserve it. So you grab the biggest project you can find, disappear into it, and slowly start falling behind. Sunil Pai, founder and CEO of Partykit, calls this the "senior engineer death spiral," and in a widely shared essay, he admits he's been caught in it more than once.

Staying afloat. The dangerous part is how normal it looks. Your standup updates stay upbeat, but the work isn't moving. You tell yourself you'll catch up tonight, then end up sacrificing sleep, meals, and time with people you love. On top of this, remote work and coding agents can make the spiral harder to spot. There's less casual contact, more individual ownership, and fewer obvious signs that someone is stuck. By the time a manager notices what’s going on, you may already be drowning.

The way out runs backward. Pai's advice is counterintuitive:

Trust ships before code. Small wins bring back momentum, and momentum brings back trust. At the end of the day, engineering is built on reputation. Read the full essay here. 

Most voice AI still works best in English.

Soniox brings speech-to-text, text-to-speech, and translation together in one real-time platform across 60+ languages, built for accurate input, natural output, and seamless multilingual conversation.

Powering real voice products at Perplexity, Samsung, Wispr Flow, Vapi, Retell, and more.

Build with Soniox

How to run agents 24/7 on your own server: This tutorial shows how to put DeepSeek Harness on a cheap VPS so your agents keep working even when your laptop is closed. You’ll learn how to connect different models through OpenRouter, switch between them mid-conversation, add plugins and custom commands, and keep long-running agent workflows online without tying them to your local machine.

TypeSafe Agent Skills (1.7K⭐️): Add Jev-style typed decisions and probabilities to Claude Code and other agents without wiring everything from scratch. The skill helps your agent design TypeSafe workflows, pull in the right docs and cookbooks, and use System One for fast routing, scoring, and other structured decisions.

How to stop AI slop before it hits prod: This guide shows how to review every PR against live production telemetry, not just tests and static checks. The idea is to predict how a code change could behave in the real system, flag risky scenarios before merge, and block changes that could break production.

Google's Addy Osmani shared a built-in Claude Code command most plugin authors miss. You write a skill or plugin, it feels like it helps, but you have no numbers to prove it.

You get a per-case score with and without the plugin, the delta, and the cost. Every eval run is a real model call billed against your plan or API account. Full case format and CI setup are in the docs.

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

Half a month into a new job, this engineer's rant about what's actually happening inside big tech is going mega viral.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ engineers and 150K+ followers on socials. Get in touch.

We put together a few guides on coding agents, agentic engineering, and leadership frameworks to help you level up in your career. Browse all our guides.

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

Until next time — The Code team