fable 5 is now permanent in max and team premium — what changed and what's still limited

Anthropic announced on July 18 and began rolling out on July 20: Claude Fable 5 is now permanently included in all Max and Team Premium subscription plans. no cutoff date, no one-time credit extension, no waiting for the next capacity announcement. if you're on Max or Team Premium, Fable 5 is just in your plan.

the caveat: at 50% of the weekly usage limits that apply to Claude Opus 5 and Sonnet 5. Anthropic hasn't published the exact weekly token numbers, but the practical implication is that heavy Fable 5 usage — long agentic runs, multi-file refactors, extended reasoning chains — will hit the cap faster than equivalent work with Opus 5.

what about pro and team standard

Pro and Team Standard users don't get the same treatment. Fable 5 remains accessible on those plans, but only through usage credits rather than the base subscription. Anthropic is providing a one-time $100 usage credit to those users as a transition bridge — enough for a substantial amount of frontier model use before the credits run out.

the practical gap this creates: Max is $100/month, Pro is $20/month. the $80 price difference now buys you permanent Fable 5 access at 50% of limits, versus credits that deplete. if you use Fable 5 regularly for complex tasks, the math will push toward Max relatively quickly. that is presumably the intent.

why now — the capacity story

Fable 5's subscription saga since June has been a capacity problem wearing the shape of a policy problem. the model launched June 9, got suspended June 12 under an export-control order, returned July 1, and then bounced in and out of subscription access as Anthropic extended the inclusion window several times while managing demand spikes.

the access pattern that developed was revealing: Anthropic kept extending subscription inclusion rather than cutting it off permanently, which meant the constraint wasn't demand (there was plenty) — it was inference capacity. each extension was Anthropic buying time for infrastructure to catch up.

two things broke the cycle. the first was the SpaceX Colossus 1 compute deal: over 300 megawatts and 220,000+ NVIDIA GPUs added to Anthropic's available infrastructure. the second was a set of efficiency improvements that reduced the per-call compute cost of running Fable 5 inference. together, they made permanent inclusion at 50% of limits feasible without degrading Max-tier response times.

what 50% of limits actually means for coding work

for typical AI coding use — asking Fable 5 to review a PR, debug a tricky bug, plan a refactor — 50% of weekly limits is unlikely to be a practical constraint. the use case where it matters is sustained agentic work: long Claude Code sessions running overnight, large-scale codebase analysis, or rapid multi-run eval cycles using Fable 5 as the judge.

for those workloads, the limit may become visible. the sensible approach is the same one that's always made sense for frontier model use: use Opus 5 or Sonnet 5 for the bulk of agentic throughput, and reach for Fable 5 specifically for the tasks where frontier reasoning is the differentiator.

the context claude.ai users should know

Fable 5 and Mythos 5 are distinct models. Fable 5 is the subscription-accessible version. Mythos 5 — the underlying model without Fable's additional safety classifiers in high-risk domains — remains API-only, not accessible through claude.ai plans at any tier. if you see Mythos 5 referenced in benchmarks, that performance ceiling is higher than what subscribers are running.

for the overwhelming majority of coding and reasoning tasks, Fable 5 and Mythos 5 are functionally equivalent. the classifier differences show up in cybersecurity, biology, chemistry, and model distillation tasks — not in shipping features.