Anthropic Reverses Course on Fable 5, Pro Loses Subscription Access

Anthropic reversed itself on Fable 5, walking back an earlier plan to pull the model from subscriptions entirely, per a post published July 18. Fable 5 now stays in Max and Team Premium permanently. It's metered at 50% of standard weekly limits, so every Fable token counts double against quota.

Those standard limits also shrink on July 20, when a Claude Code promotion expires. Same subscription price, smaller bucket, faster burn.

Pro and Team Standard subscribers get a worse deal. Subscription access to Fable 5 ends outright.

They get a one-time $100 usage credit, then API billing kicks in at $10 per million input tokens and $50 per million output. For a $20-a-month Pro subscriber running Fable inside Claude Code, that credit covers roughly 2 million output tokens. I've burned through that much in a week on one refactor.

The timing isn't subtle. GPT-5.6 Sol reached general availability on July 9 at $5 per million input tokens and $30 per million output. That's half Anthropic's input cost and 60% of its output cost. Pulling Fable from subscriptions two weeks after a competitor made a cheaper model generally available would have looked bad.

The real signal

The pricing table isn't the real story here. A forty-minute agent loop that reads three hundred files pulls more inference than a chat session ever will. Anthropic billed both the $20-a-month chat user and the $20-a-month agent operator at the same rate for eighteen months. That assumption is what's actually breaking.

That math should change how you allocate models, not just budget for them. Bulk implementation work, the stuff a test suite already checks, doesn't need the flagship. Fable makes sense for judgment calls: reviewing an approach, catching an edge case a linter would miss. If you're on Pro running anything on a schedule, total your monthly output-token volume now, before that $100 credit runs out. After that, the API meter doesn't care how big your cron job gets.