Fable 5 is now fully metered: $10 per million input tokens from today
If you are one of the many Pro, Max or Team subscribers who have been leaning on Claude Fable 5 for long-horizon agentic tasks, today is the day your bill changes materially. From 7 July, Fable 5 is fully decoupled from standard subscription usage limits and moves to pure metered credits at $10 per million input tokens and $50 per million output tokens, the highest rate Anthropic has ever charged for a generally available model. That is roughly double what you would pay for Claude Opus 4.8, so anyone running agentic Fable 5 sessions across large codebases should expect the meter to move quickly.
Anthropic says this is a temporary capacity management measure while it scales Fable 5's infrastructure back up after a turbulent six weeks that included an export-control suspension and a lengthy outage. The intention is to fold Fable 5 back into standard plans once capacity allows. For now, if you want to continue using it, go to Settings, then Billing, then Usage Credits on claude.ai and top up before you run into a mid-task cutoff.
Anthropic published a case study on how the Government of Alberta used Claude Code, running on Opus and Sonnet models, to scan 466 million lines of government code in 20 hours, identify security vulnerabilities and ship fixes. If you are evaluating agentic coding tools for large-scale legacy remediation in regulated environments, this is the most concrete real-world evidence yet. Anthropic Newsroom, 6 Jul →
21 new multilingual Grok voices arrive with improved naturalness across the originals
If you use Grok Voice day to day, either through the app or via the API, xAI has made it sound considerably more like a premium voice assistant should. The company shipped 21 new multilingual voices for Grok Voice alongside improved naturalness for the original five, meaning both wider accent and language coverage and a subtler, less robotic delivery on the voices you already know.
Voice remains one of the most competitive fronts in consumer AI right now, and xAI has leaned into it consistently this year: custom voice cloning, agent-building tools, and now this expansion. The lab is treating voice quality as a genuine differentiator rather than a feature bolted onto a text model. If you build voice agents or simply use Grok Voice for everyday tasks, it is worth trying the new voices directly, since naturalness improvements like this tend to be more noticeable in use than in a changelog. The update surfaced through xAI's own news page before mainstream coverage picked it up.
Today's themes
A few patterns stand out today. Anthropic's shift of Fable 5 to fully metered credits is a reminder that even the biggest labs are capacity-constrained when serving their most demanding frontier models at scale, and that tightening government controls on frontier AI keep colliding with the practical business of keeping a model running for paying customers. xAI's voice expansion, surfaced via its own news page before any mainstream coverage, shows one of the clearest fronts where labs are still competing on raw product polish rather than governance positioning. The Alberta case study, also found via Anthropic's own newsroom, points to agentic coding tools graduating from demos into genuine large-scale legacy remediation inside government: worth watching if you are evaluating these tools for infrastructure you actually depend on.