Claude Code Limits +50%, Fable 5 Lands in Max & Team

Claude Code weekly limits will stay 50% higher through August 19, 2026 — and Claude Fable 5, which was scheduled to disappear from subscription plans on July 19, is instead becoming a permanent fixture of Max and Team Premium at 50% of weekly usage limits. Anthropic shipped both changes in the space of two days, via two X posts and a Help Center update.

The asymmetry is the story. Max, Team Premium, and premium seats on legacy seat-based Enterprise plans keep Anthropic’s frontier model inside their subscription. Pro and Team Standard users lose included access entirely from July 20, receive a one-time $100 usage credit, and then pay metered rates of $10 per million input tokens and $50 per million output tokens — the widest input/output spread in Anthropic’s current lineup.

This post covers exactly what changed and where each claim comes from, the full deadline timeline (this is at minimum the fourth deadline move on this promo lineage since May), a plan-by-plan matrix of the post-July-20 terms, the real-world math on the $100 credit, and what the reversal signals about competitive pressure from GPT-5.6 Sol and aggressively priced Chinese coding plans.

There is no single press release to point at. The July news is a composite of three artifacts, and they carry different pieces of the story. Around July 17, Anthropic’s official @claudeai account posted that, beginning July 20, Fable 5 would be included in all Max and Team Premium plans at 50% of limits, with Pro and Team Standard users continuing via usage credits plus a one-time $100 credit. On July 18, the @ClaudeDevs account added the Claude Code piece: weekly limits stay 50% higher, now through August 19, for all Pro, Max, Team, and seat-based Enterprise users.

The third artifact is the one you can actually load: the Claude Help Center article “Claude Fable 5 on your plan.” It confirms the Max / Team Premium 50%-of-limits mechanic, the July 19 11:59:59 PM PT end of the promotional window, the July 20 effective date, and the requirement that Fable 5 in Claude Code needs version 2.1.170 or later. Notably, that article says nothing about Claude Code weekly limits or August 19 — the two announcements travel on separate sourcing chains, and coverage that blends them into one “official update” is glossing over the difference.

Anthropic’s stated reason for the staged rollout, via the X post as reported by TechTimes: demand for Fable 5 has been challenging to predict, requiring incremental capacity build-out since the model’s June 9 launch. Our July 13 piece on the rolling access uncertainty — written when the July 19 cliff was still the live deadline — treated that unpredictability as a vendor risk to hedge. This announcement resolves the specific uncertainty it covered, in a direction few predicted: not another extension of free access, but permanent-but-reduced terms.

Most coverage treats each of these announcements as a standalone news item. Read together, they form a single lineage: a limit boost and an included-access window that Anthropic has now extended, re-extended, and finally restructured — at minimum the fourth deadline move since May 2026. The table below is the full chronology, with what each announcement changed and who it covered.

Two of those rows have their own coverage on this site, written as the news landed: our July 7 post on the first extension covered the move to July 12 as it stood at the time, and our July 7 usage-credits pricing guide documented the $10/$50 credit mechanics under the earlier all-paid-plans terms — both now superseded on the deadline specifics by the July 20 restructuring described here.

The interpretive read: a deadline that moves four times in ten weeks is not a deadline, it is a pricing experiment running in production. Each extension bought Anthropic another observation window on what Fable 5 demand actually looks like under subscription terms, and the final shape — permanent access for the highest-paying tiers, metered access below — is what you converge on when the data says the top tiers retain because of the model and the lower tiers cost more to serve than they pay. For teams, the practical lesson cuts the other way: any AI-tool budget built on a promotional deadline should assume the deadline is soft, in both directions.

Existing coverage states the new rules tier by tier in prose. Here is the same information as a single side-by-side matrix — who keeps included Fable 5 access, who moves to usage credits, who gets the $100 cushion, and who is covered by the Claude Code limit boost through August 19.

Two caveats on the matrix. Consumption-based Enterprise seats were explicitly excluded from the July 13 extension leg per Help Net Security’s reporting, and nothing in the July 18 announcement names them — assume excluded until Anthropic says otherwise. And the earlier extension legs applied automatically with no user action required; since the August 19 leg is a continuation of the same program rather than a new one, the same auto-apply mechanic is reasonable to expect, but Anthropic has not separately reconfirmed it for this leg.

The one-time $100 credit for Pro and Team Standard users sounds generous until you run it against Fable 5's published rates. At $50 per million output tokens, $100 buys exactly 2 million output tokens. TechTimes’ analysis made the point first: a single output-heavy session — around 2 million output tokens, which it characterizes as modest for an autonomous multi-file coding migration — exhausts the entire credit. That is an estimate built on published pricing, not a vendor-stated conversion, and it ignores input-token costs entirely, which makes it a generous floor rather than a forecast.

Extending the same output-only arithmetic to smaller sessions: a session emitting roughly 100K output tokens costs $5 at the published rate, so the credit covers about twenty of them; at roughly 250K output tokens — a feature build with tests — each session runs $12.50, or about eight sessions per credit. These session profiles are illustrative assumptions, not measurements, and real bills will be higher once input tokens are counted. Fable 5's spec sheet explains why the output side dominates: the model supports up to 128K output tokens per response against a 1M-token context window, per Anthropic’s launch documentation, so long agentic runs are output-heavy by design.

One report has floated more specific credit terms — a per-seat structure for Team plans with an organization cap and expiry dates — but none of those sub-figures appear in the Help Center article or any independently corroborated source we could verify, so we are not repeating them as fact. The verified shape is simpler: a one-time $100 credit for eligible Pro and Team Standard users, then metered rates. We run this blog’s entire content pipeline on Fable 5 in Claude Code, so this is math we watch operationally: for heavy agentic use, $100 is a runway measured in days, not months.

The permanent-access reversal did not happen in a vacuum. Simon Willison — whose framing has been widely echoed — reports that Anthropic’s original plan was to drop Fable 5 from subscriptions entirely and move it to metered-only pricing, and argues the competitive landscape made that untenable. That is his interpretation, not an Anthropic statement. But the numbers underneath it are published: OpenAI’s GPT-5.6 Sol reached general availability on July 9 at $5/$30 per Mtok — half Fable 5's input rate and 60% of its output rate — while scoring within a point of Fable 5 on the Artificial Analysis Intelligence Index (58.9 vs 59.9, as of July 18, via TechTimes).

The per-task picture is harsher than the per-token one. Artificial Analysis found GPT-5.6 Sol completes equivalent tasks at roughly one-third Fable 5's per-task cost, because it emits fewer output tokens for comparable results — a dynamic we examined in detail in our GPT-5.6 Sol vs Fable 5 price and access comparison. The pressure from below is real too: Kimi’s frontier models and aggressively priced Chinese coding plans that explicitly market Claude Code compatibility keep resetting the floor, as we covered in the Kimi K3 vs Fable 5 frontier comparison. Willison goes one step further and speculates Anthropic may need to dial back training efforts to free GPU capacity for serving Fable 5 to the wider subscriber base — clearly labeled speculation on his part, worth knowing and not worth repeating as fact.

The Princeton AI Newsletter’s Tiger VC memo distills the strategic read: frontier access is no longer unlimited at the top tier, and labs are now competing on access economics, not just capability benchmarks. That matches the shape of this announcement precisely — Anthropic did not cut Fable 5's price to meet GPT-5.6 Sol; it restructured who gets the model inside a subscription and how much of it.

The new terms make plan choice a genuine decision rather than a default. The matrix below is our recommendation set for the post-July-20 landscape, based on the verified terms and the credit math above.

For the routing question — which Claude model should carry which workload now that Fable 5 access is asymmetric — our Sonnet 5 vs Opus 4.8 vs Fable 5 decision matrix holds up well under the new terms, and the subscriptions-vs-usage-credits framework is the right conceptual lens for the Pro-vs-Max crossover decision. Looking forward, expect more of this, not less: when near-frontier capability is available at a third of the per-task cost, every lab’s premium tier has to justify itself on access terms, and those terms will keep being renegotiated mid-cycle. If your team is building processes on top of these tools and needs the vendor-risk side handled — model routing, cost ceilings, fallback plans — that is exactly the ground our AI transformation engagements cover.