Not a review and not a changelog. Each guide is the set of decisions an operator has to make before the model runs in production: what it costs, what to route to it, how to govern it, and when to say no.
Identical pricing to Claude Fable turned a procurement decision into a routing decision. This guide covers the pricing table nobody read past the first line, the harness that moved the score more than the model did, and the five questions to answer before it runs unattended.
Persistent agents with your credentials on a shared machine, at $300 a month. The three rules that make them work were found twice, independently, by practitioners and by the vendor's own docs. Everything else in this guide is downstream of those three.
The guide the other two are written against. Fable 5 is where the operating model changed: from supervising steps to directing outcomes, from briefing to interviewing, from watching every step to handling the exceptions. Start here if you are starting anywhere.
Not a guide. A showcase. Twenty things people built with GPT-6 Astra in the seven days after launch, collected with links to the originals, followed by the part the timeline skipped: what the twelve actually prove.
Not guides. The two pages that make the rest of the Playbook navigable: every benchmark score we have tracked, and every term a vendor deck assumes you already know.