After weeks of uncertainty, Anthropic is giving its highest-paying subscribers a permanent—if limited—home for its most advanced public AI model. Starting July 20, Claude Fable 5 will be included in all Max and Team Premium plans, but with access capped at 50 percent of those plans’ normal weekly usage limits. Pro and Team Standard customers won’t get it in their core allowance at all; they’ll pay for every interaction through meter credits.

The change, announced by the company on Saturday, sets a new baseline for who gets Fable 5 and how much they can use it before extra charges kick in. It also closes a chapter of stopgap extensions and plan flipping that began when the model was abruptly disabled in June over U.S. export-control concerns.

The new Fable 5 deal

Anthropic’s July 18 post clarified what the model will cost different subscriber tiers starting July 20:

  • Max and Team Premium: Fable 5 returns as an included perk, but usage is limited to 50% of the plan’s weekly message limit. The company called this “standardising access” for the plans that lean most heavily on the model. If you burn through that half, you’ll keep chatting with Claude’s other models—or start eating into usage credits.
  • Pro and Team Standard: Fable 5 stays credit-only. That means no free weekly allocation; every prompt to the model draws from a purchased pool of credits, billed at Anthropic’s standard overage rates. To soften the blow, the company is giving these users a one-time $100 credit.

This split isn’t entirely new. Fable 5 debuted in paid plans on July 1 at up to half of limits, but that temporary window was supposed to end July 7. Instead, Anthropic kept extending access while it struggled to bring enough compute online. Now the temporary arrangement becomes permanent—except for lower-tier subscribers, who were cut loose from the included allowance entirely.

Who gets what—and what it costs

For Windows users running Claude in the desktop app, browser, or the Claude Code command-line tool, the model itself isn’t disappearing. The difference is entitlement: some pay for it upfront as part of their subscription, others pay as they go.

Max subscribers paid for the highest-tier plans ($100/month for Max 5x, $200/month for Max 20x) gain the most predictability. Their Fable 5 access is ring-fenced at 50%—not generous, but a known quantity. They can plan projects around that ceiling.

Team Premium ($30/user/month) gets the same half-limit inclusion, with the added wrinkle that administrators can set organization-wide credit-spend caps. That’s useful for controlling costs, but it also means a developer’s Fable 5 consumption can be throttled by a centrally managed budget, not just their own usage.

Pro ($20/month) and Team Standard ($15/user/month) subscribers now must treat Fable 5 as a metered premium feature. The one-time $100 credit is a helpful buffer, but it won’t last long for anyone who uses the model as their daily driver. At Anthropic’s standard credit rates, 100 dollars might buy a few days of heavy coding or a handful of long-context analysis sessions.

Heavy users feel the squeeze

Fable 5 was billed by Anthropic as its most capable public model, with strengths in software engineering, long-document reasoning, vision tasks, and agent-style work. Those are exactly the workloads that gobble up message allowances. A single debug session with Claude Code can generate dozens of prompt-response pairs, making the 50% cap feel restrictive.

For developers who treat Fable 5 as their default assistant, the new structure forces a hard re-think. You can’t just fire off every question to the most powerful model. You’ll need to triage: reserve Fable for the truly gnarly problems, and offload routine prompts to Claude Sonnet or other less expensive models.

Windows users who rely on Claude Code for multi-file edits or automated pull-request reviews should pay particular attention. If your workflow scripts assume Fable 5 is always available under a subscription limit, it will break once that allowance runs dry mid-week. Check your automation, and consider adding fallback logic that scales down to a cheaper model when credits kick in.

From suspension to staged return

To understand why Anthropic is metering its best model this way, you have to look back a few months.

Claude Fable 5 is the public version of Mythos 5, first shown to the world in early June. It was immediately slammed by demand. Then, less than three weeks later, both Mythos 5 and Fable 5 were disabled for all users after the U.S. government restricted access for foreign nationals, citing national security concerns tied to advanced AI.

When the models returned on July 1, Anthropic said it would include Fable 5 in several paid plans through July 7, at up to half of usage limits. That was always described as a temporary, capacity-exploring measure. But as demand proved “challenging to predict,” the company issued multiple extensions while it scrambled to bring more compute online.

Startup Fortune characterized the situation bluntly: Anthropic was running out of compute, not tweaking a subscription tier for fun. The company itself hasn’t published detailed infrastructure figures, but its support documentation makes clear that usage credits are designed to let paid subscribers keep using Claude after included limits—a revenue model that helps smooth out demand spikes.

The July 20 plan is a stabilization effort. Instead of indefinite temporary windows, Anthropic is drawing a line: premium plans get a permanent but capped allowance; everyone else pays per use. The company said it will “continue investing in additional capacity,” but for now, this is the new normal.

What you should do before Sunday

The new rules take effect July 20. Here’s a practical punch list for anyone using Fable 5 on Windows—whether you’re a solo dev or an IT admin managing a team:

  • Know your plan. Log into the Claude web dashboard and confirm whether your account is on Max, Team Premium, Pro, or Team Standard. This determines your inclusion vs. credits status.
  • Review credit settings. If you’re on Pro or Team Standard, ensure usage credits are enabled and that you’ve set a comfortable monthly spend cap. The one-time $100 credit is automatic, but after it’s gone, you’ll start racking up charges unless you cap them.
  • Reserve Fable 5 for heavy lifting. Route everyday questions, drafts, and simple code completions to Claude Sonnet or cheaper models. Use Fable only when you genuinely need deep reasoning, multi-file context, or agentic decision-making.
  • Audit your automations and scripts. If you’ve hard-coded Fable 5 as the model in Claude Code workflows, CI/CD pipelines, or research scripts, switch them to dynamic selection. Have them fall back to a less expensive model when credits are active or when the 50% allowance is exhausted.
  • For team admins: Check spend controls. Team Premium admins have organizational credit limits. Make sure the cap is high enough to support critical work but low enough to avoid bill shock. Communicate the new Fable 5 policy to your team explicitly—especially to any developers who used it heavily during the free-trial extensions.

These aren’t speculative suggestions. Anthropic’s own documentation confirms that after included limits, consumption is billed at standard rates with no warning beyond the dashboard.

Looking ahead

Anthropic says it will keep investing in infrastructure, but it hasn’t hinted at when—or if—the 50% cap might rise. The company is clearly capacity-constrained, and the pressure will only increase as more users discover Fable 5’s coding and analysis chops. For now, the message is plain: Fable 5 is a premium resource, and you need to spend it wisely. The July 20 changes remove the short-term confusion, but they also put every subscriber on notice that unlimited access to cutting-edge AI isn’t coming cheap—or anytime soon.