Loading today’s stories
Product1 distinct publisher3 min readPublished
Pro and standard Team seats have to buy usage credits to run Fable 5.1, Max seats may spend half a weekly limit on it, and those rules live in the help center, not the pricing page.
The Product Desk · Product desk
leadership
Anthropic cuts Fable 5.1 prices by 25% and launches two-tier safeguard system with Mythos 5.12 distinct publishers
invest
Anthropic Cuts Cache-Read Prices by 75%; Cache Reads Were ~60% of a Heavy Agent's Bill Before the Cut1 distinct publisher
build
Fable 5.1's 52.6% science score arrives on a benchmark that was five days old1 distinct publisher
product
Four Claude models, four surfaces, one incident: tier fallback is inside the blast radius1 distinct publisher
Compiled by The Product DeskSomething wrong?How this is made
Anthropic's FAQ heads off the reading that matters for capacity planning: on Max and on premium Team and Enterprise seats, the Fable share is half of the weekly limit, drawn from the same pool every other model uses, not half added on top [7]. The same help article says Fable spends that pool faster than other Claude models [8]. Half an allowance, drawn at a higher rate, buys less than half a week of work, and the ceiling arrives earlier than the seat price implies.
The defaults tilt the same way. Fable 5.1 opens at High effort in Claude Code and at Medium in Claude Cowork and on claude.ai [9], and Anthropic's article on effort levels says the higher setting is more thorough and reaches the limit sooner [10]. The same person on the same seat gets a thriftier model in the browser than in the terminal; the control sits in the model menu next to the send button, and thinking cannot be turned off at any level [9][11].
The discount that got the headlines lands on a different bill. It comes from cache reads, now 75 percent cheaper at $0.25 per million tokens [13], which puts the pre-cut price at $1.00 per million, since $0.25 is the surviving quarter [1]. Base rates did not move, at $10 per million input tokens and $50 per million output [14], so one output token still costs what 200 cache reads cost [2]. Subscription prices are untouched by any of it [15]. Whether a workload sees 25 or 45 percent depends on how much of it is cache hits, and a Pro seat sees none of it, because a Pro seat is not billed per token [12][3].
Capability did move, and the claims are checkable. Anthropic says Claude Code users should see roughly 60 percent fewer interventions per session, and that safeguards fire 85 percent less often on harmless biology and medicine questions than at Fable 5's launch [20]; the model may now identify software vulnerabilities while remaining barred from building exploits [19]. Mythos 5.1, which Anthropic describes as the same model with different safeguards, stays inside two verification programs currently open only to US organizations [18].
For a rollout, build an entitlement table. One row per tier you actually buy (free, Pro or standard Team, Max or premium seat) and two columns: may this seat run Fable at all, and what happens at the cap [3]. Add one more column for the fallback: the model the work drops to for the rest of the week. Anthropic's own answer for Pro is usage credits, an upgrade to Max, or Opus 5, which stays inside the allowance [16]. If the team spends Thursday and Friday on the fallback, the fallback is the model you bought, and that entitlement table is what your finance partner needs to see.
Ranked by verification strength, evidence, and original report placement.
Anthropic launched Claude Fable 5.1 on Monday and calls it its most advanced model for coding and knowledge work; paying customers see it in the model picker right away, marked with a "New" badge.
Free accounts do not get Fable at all, since Fable is limited to paid plans.
On Pro plans and on standard seats of Team plans, Fable 5 and Fable 5.1 are not included in the plan's usage limits; anyone who wants to use them there pays through usage credits, on top of the 20 dollars a month.
When Fable 5 moved to the usage-credit model in July, affected Pro customers received a one-time credit; for Fable 5.1 the help article states explicitly that there is no equivalent credit.
At launch on Monday, Max customers opening claude.ai saw a notice window saying Fable 5.1 was included for up to 50 percent of their Max allowance; that number appears in neither the announcement post nor the pricing page, and the help center explains it.
On Max plans and on premium seats of Team and Enterprise plans, a customer may spend at most half of the weekly limit on Fable models without paying extra; after that the options are to continue with usage credits or switch back to a different model.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Vendor documentation, read by one outlet
The entitlement facts come from Anthropic's own help articles and FAQ, which is the best available source for what a company charges and also the kind of page that can be rewritten without notice. What is absent is a second reader: no other publication in our coverage checked the same pages, no revision is dated or linked, and the pivotal 50 percent Max figure reached customers through a launch-day notice window rather than any durable published document.
Shipped to every paid seat, used by nobody we can count
Deployment is not in question: the model appeared in paying customers' pickers on day one and the metering rules applied immediately. Uptake is entirely unobserved. Nothing here counts seats, credit purchases, or Max users who ran into the half-week wall, and the reduction figures Anthropic quotes for safeguards are its own measurements rather than field data.
The savings are Anthropic's arithmetic, not anyone's measurement
Two headline percentages — 25 percent cheaper typically, up to 45 percent for agentic work — are the vendor's own numbers derived from its own cache-read discount, and they land only on per-token API bills. Notebookcheck's contribution is to mark exactly where they stop, and the reporting stays inside quotable documentation rather than adding claims of its own. The overstatement that remains belongs to the launch: a model presented as the most advanced arrives in a picker where the largest group of paying customers cannot run it without paying a second time.
Metering revenue on one side, subscriber grievance on the other
Anthropic's incentive is structural: billing the most expensive model separately from the plan turns it from a margin problem into a revenue line, and the rules for doing so live in a support article rather than on the page a buyer reads before subscribing. Declining to repeat July's one-time credit points the same way. Notebookcheck's incentive is narrower and visible in the framing — a consumer-tech outlet finds its audience among subscribers who feel shortchanged, which is why the piece reaches for stolen Claude sessions as an illustration of a drained allowance.
Mechanics solid, magnitudes unknown
What each seat is entitled to is checkable against Anthropic's own pages and stays internally consistent across every section of this reporting, including the FAQ line that pre-empts the additive misreading. How much faster Fable drains a weekly limit, how many customers actually hit the halfway mark, and whether July's credit set any expectation at all are all unestablished — and one outlet's reading of support documentation stands behind the whole account.