Tuesday, July 7th, 2026: If you’re on Claude Pro (the $20/month plan), Claude Max (the $100/month plan), or Claude Team ($25/user/month for Standard, $150/user/month for Premium), it’s your last day to use Claude Fable 5 bundled within your existing subscription limits. Starting tomorrow, you’ll need metered usage credits to get your paws on that sweet super-inference.
I suspect a lot of power users are going to be mainlining Fable 5 today; I myself will be availing myself of it via NetFoundry’s team plan (hey, if a company offers a perk, you use it!).
If you’re having trouble coming up with ideas for what to do with Fable 5 on this last day, I have suggestions:
- Top (and obvious) priority: Use it to bring whatever Claude 5-based work to a state where you can hand if off to another model or human.
- Treat it like the member of the group project who did all the work and is now moving far away. Have it walk you through any code that you asked it to generate that you don’t understand.
- Devote a day a week to sharpening a skill that you’ve been outsourcing to Fable 5. Not “donkey work,” but an actual skill that you don’t have (yet).
- Run the same hard task on Fable 5 and Opus 4.8 side by side. You’ll have a much better idea whether $10/$50-per-million-token price tag is justified for your workloads.
In the spirit of “meta”, I also asked Fable 5 what one should do on the last “free” day of Fable 5. Its answers:
- Feed it the scariest codebase you have access to. The multi-repo, “the guy who wrote this left in 2019” kind. Fable 5’s benchmark edge is agentic coding and security audits, so this is the one day you can get a frontier-tier code review without doing token math in your head.
- Run one genuinely long agentic task end to end. Not “write me a function” — more like “plan, scaffold, build, and test this MVP.” Long-horizon work is the whole point of the Mythos tier; short tasks reportedly come out nearly identical to Opus 4.8, which is like renting a Ferrari to drive to the mailbox.
- Do the research brief for a decision you’ve been avoiding. Tech stack choice, framework migration, “should we self-host this” — anything with high decision value, lots of context, and many moving parts. That’s where the extra capability actually shows up.
- Audit your docs the way an AI reads them. Hand it your entire documentation set and ask what an LLM would get wrong, miss, or hallucinate about your product.
- Throw a big, ugly dataset at it. The CSV graveyard you’ve been meaning to analyze since Q1. Dense, messy, multi-step data work is squarely in its wheelhouse.
I also asked for some of the answers to be impractical and humorous. Its answers:
- Ask it to explain the month it just had. Launched June 9, hit with a US export-control directive on June 12, suspended globally within 90 minutes, cleared and restored July 1. Few software products have a redemption arc; fewer can narrate their own.
- Use the most expensive generally available AI model in history to write your grocery list. Pure decadence. The token cost of “eggs, milk, jalapeños” has never been higher, and it never will be again. Probably.
- Have it arrange “Master of Puppets” for solo accordion. Is this a good use of Mythos-class reasoning? No. Will I do it convincingly? Also possibly no. But it’s free until midnight. [This one seems to have been aimed directly at me.]
- Ask Fable 5 itself when it’s coming back to subscriptions, then screenshot its diplomatic non-answer for posterity. (Anthropic’s actual answer: “when sufficient capacity allows.” My answer would be roughly the same, but with more charm.)
Happy Fable 5 Day!