🧠 Locked Out 🔒, Leveled Up

OpenAI previewed three new models: Sol for the hardest problems like complex coding and security, Terra for everyday business work at half the cost of the older 5.5, and Luna for fast, cheap, high-volume tasks.

Here's the twist. At the US government's request, Sol is rolling out as a limited preview to roughly 20 approved partners instead of the open launch OpenAI planned. It's the first time a US lab has shipped a frontier model through a government-managed access list. General availability is promised "in the coming weeks."

When it opens up, pricing is set at $5 in and $30 out per million tokens for Sol, half that for Terra, and $1 and $6 for Luna. OpenAI said plainly that it doesn't think this kind of gatekeeping should become the norm.

Boris Cherny, who built Claude Code, shared a sharp reflection on how roles are blurring as AI reshapes how teams work. He sees five archetypes: the Prototyper who fires off ideas that mostly don't ship, the Builder who turns an idea into a real product, the Sweeper who simplifies and optimizes, the Grower who iterates toward product-market fit, and the Maintainer who keeps mature systems reliable at scale.

His point lands for anyone running lean: these archetypes aren't tied to job function. A designer, a PM, and an engineer can each be any of them. If you're a solopreneur, you're probably wearing all five hats already, and naming them is a surprisingly useful way to see where your week actually goes.

xAI's Grok 4.5 is built on its new 1.5-trillion-parameter V9 model, with coding data from Cursor mixed into training. For now it lives only in private beta at SpaceX and Tesla.

Musk says early internal evals put it close to or past Anthropic's Opus, though those are his own numbers and not independent benchmarks yet. He also claims SpaceX will ship a model trained from scratch every single month for the rest of the year, which would be a brutal release pace if it holds.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.

If you use Claude Code (Anthropic's command-line coding agent) and you're tired of babysitting one task at a time, here's the trick: run four agents side by side. The setup uses tmux, a free terminal tool that splits one screen into panes, so each Claude works a different job while you watch all four at once. One writes tests, one refactors, one hunts a bug, one writes the docs. Here's the full walkthrough:

2. 💸 Get way more out of your Claude subscription

A few small habits stretch your plan a long way. The biggest one: don't let a session go cold. Claude only caches your context for a few minutes, so reopening an old conversation after a break forces it to re-read everything and quietly torches your limit. Before you stop, ask Claude to write a HANDOFF.md file summarizing the project, your key decisions, and what's next, then start a fresh session by feeding it that file. Match the model to the job too: Haiku for brainstorming, Sonnet 4.6 for most real work, and Opus 4.8 only for heavy planning. And one clear, batched prompt beats ten tiny back-and-forth messages every time.

If you still haven’t tapped into Claude connectors, you really are missing out. These three are worth turning on this week:

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.