As Simon Willison details, the core news here is arguably not the model's performance but its economics and regulatory positioning. The effective price increase via tokenization, while buried in technical documentation, represents a subtle but significant shift in how AI vendors can adjust unit economics without changing headline rates. This move could signal a broader industry trend toward optimizing revenue through architectural changes rather than transparent pricing adjustments.
Meanwhile, the explicit framing around reduced cyber capabilities to satisfy U.S. government concerns underscores how model cards are becoming as much a compliance document as a technical one, potentially foreshadowing future releases where capability throttling is a primary feature.