Two companies dominated the last two weeks, and with a ton of announcements, there’s one major theme: the models keep getting better (and the way you pay for them is changing under our feet).
Fable is back, and it comes with a meter
Last month, I covered Claude Fable 5's launch, the first publicly available version of Anthropic's Mythos-class models, and told most of you to leave it on the shelf. Three days later, the US Department of Commerce made that mandatory, ordering it pulled worldwide (their reasons were a little more serious than mine). The export controls were lifted on July 1, and Fable is back for everybody.
This thing can work for 10+ hours at a time on a single task, build computer games in one shot, and read every email you've ever written and received faster than you can delete today's spam. It's built for long, complex projects, the kind that run for hours or days inside agent tools while the model plans, delegates, and checks its own work. During early testing, Stripe told Anthropic the model compressed months of engineering into days, handling a codebase migration in a day that would have taken a team over two months by hand. But that also means my advice from the last issue still stands: it's not for everyday tasks - drafting, analysis, meeting prep, a quick check on a spreadsheet… you should keep using Opus and Sonnet for those. Otherwise, Fable is like using a Formula 1 car for a grocery run.
Now the meter (not the European cousin of the yard, meter like your electric meter at home). Through this Sunday, July 12, Claude Pro, Max, Team, and some Enterprise subscribers can use Fable for up to half their weekly plan limits at no extra cost. After that, Fable runs on usage credits purchased on top of your subscription. That means your AI line item is starting to look less like a software license and more like a utility bill. And for more clients, this will be confusing for a while.
One more lesson from the three-week blackout. Back in April, during the Claude compute crunch, I encouraged you to have a backup plan instead of building your whole operation around a single provider. The Commerce Department just made my argument for me. If a workflow that matters to your business is hard-wired to one specific model, you now know it can vanish overnight, and not because of anything you did.
OpenAI's answer: Sol, Terra, and Luna
Eight days after Fable came back, OpenAI made GPT-5.6 generally available and retired its old naming mess in the process. The number now tells you the generation, and the name tells you the tier (the sun, the earth, and the moon, in case your Latin is rusty):
- Sol is the flagship for the hardest work
- Terra is the everyday workhorse. OpenAI says it performs on par with GPT-5.5 at a lower cost
- Luna is the fast, cheap tier for routine, high-volume work.
If you subscribe to ChatGPT, there's nothing to install. Sol now powers the reasoning options on eligible paid plans, and the lower tiers show up in Codex and the new Work product. OpenAI's launch charts show Sol beating Fable on several coding benchmarks at a lower cost, while Fable still leads on others, so I'd hold the scoreboard loosely. The takeaway is the naming system itself. Matching the size of the model to the size of the job no longer requires an advanced degree. And for our clients who are wrestling with usage limits, resetting usage, and new metered billing - choosing the right model for the task makes all the difference.
Codex grew up and got a business card
The same day (literally yesterday), OpenAI launched ChatGPT Work, an agent built on Codex, the coding tool it originally shipped for developers. OpenAI says more than a million people were using Codex for work that had nothing to do with code (I’m absolutely one of them), so they followed the users. Work connects to Slack, Teams, Google Drive, SharePoint, email, calendars, and CRMs, then produces finished output: spreadsheets, decks, documents, even simple web apps. It can stay on a project for hours by breaking it into steps, and the Codex desktop app is merging into a single ChatGPT app with Chat, Work, and Codex side by side. That may be a lot to swallow, so let me encourage you simply: ChatGPT Work should 100% replace ChatGPT in your life. It does everything ChatGPT did and significantly more. The closest I can come to describing it is trying to tell you in 2007 that you can retire your Motorola RAZR phone because this new thing from Apple can make calls/texts… and a whole lot more.
If that road sounds familiar, Anthropic paved it first when Claude Code grew into Cowork. Both labs watched knowledge workers adopt their developer tools and concluded that agents doing multi-step work across your apps are the next product for everyone, not just engineers. It almost feels like this is a funeral for Chatbots - the question is how long will it take for the average person to figure that out.
Rollout notes, because I know the first question: on web and mobile, Work reaches Pro, Enterprise, and Edu plans first, with Plus and Business following over the coming days, while the desktop app is available on every plan. Admins control which plugins and permissions are enabled, so if your team can't see it yet, nothing is broken. It's a rollout or permissions thing, and we're glad to help you sort it. Billing here is usage-based too, drawing from the same consumption pool as Codex, which makes it two for two on the metering trend.
When you pay for capability by the gallon, spraying everything gets expensive, and your aim (or AIm) is going to be scrutinized more. “Why is so and so in finance using 10x the AI credits as everyone else” is a more common question than you’d realize these days. If you want a second set of eyes on which of this belongs in your budget and which of it is noise for your team, reply to this email and we'll talk it through.