/Tech13h ago

Anthropic releases Claude Fable 5, using dynamic classifier gates to route sensitive requests to Opus 4.8

Story Overview

Anthropic has opened access to a Mythos-class model for ordinary users by releasing Claude Fable 5, which keeps most queries on the full system while routing anything flagged as sensitive in cybersecurity, biology, or model-copying domains straight to the more restricted Opus 4.8.

1693.3K137283748.8K
Cristian Garcia
Rohan Paul
Lisan al Gaib
Danielle Fong 🔆
elvis#510
Original post
Rohan Paul@rohanpaul_ai

Today’s edition of my newsletter just went out.

🔗 https://www.rohan-paul.com/p/anthropic-finally-released-claude

🗞️ Claude’s ‘too dangerous’ AI model is finally public. But there’s a catch

🗞️ Cognition is introducing FrontierCode, a coding benchmark built to test whether AI code is good enough for a real maintainer to merge, not just whether it passes tests.

🗞️ This is the silent limiter on Claude Fable 5 - It cannot be used for really advanced AI research stuff.

🗞️ New Anthropic research shows AI agents may look brilliant at code, but in biology they can fail before the science starts.

🗞️ Very useful recommendation for pushing Claude Code to its full potential. by Thariq, from Claude Code team.

2:43 PM · Jun 9, 2026 · 3K Views
Pricing Watch

Promotional access ends soon on paid plans

Fable 5 lands immediately on Pro, Team, Enterprise, and major cloud platforms at $10 per million input tokens and $50 per million output, included free through June 22 before shifting to usage billing.

Developer Impact

Classifier gates replace blunt refusals

Dynamic detection hands off risky requests instead of blocking them outright, with users notified on fallback and over 95 percent of sessions expected to stay on Fable 5, though real-world trigger rates remain unquantified so far.

Sentiment

Many users criticized Anthropic's Claude Fable 5 release as pointless or overly restricted by safety classifiers and refusals, while others praised its low exploit rates and strong performance on allowed tasks.

Pos
33.3%
Neg
66.7%
10 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS11.6K
Sarah Yang@sarahyang_00

@steph_palazzolo Anthropic is pulling a classic luxury car pricing anchor move by leaking a 5x price tag first so customers feel like they are getting a massive discount when they pay 2x for a neutered Fable

1dViews 11.6KLikes 51Bookmarks 2
BOOKMARKS7LIKES122REPLIES9
Lisan al Gaib@scaling01

I know a thing that I'm actually number 1 in the world

I'm 2/2 in reporting first (minutes before the official launch) about 10T+ model system cards

I was first for GPT-4.5 and for Claude Fable 5

Lisan al Gaib@scaling01

Claude Mythos & Claude Fable System Card

7hViews 8.4KLikes 122Bookmarks 7
RETWEETS22
lucy 🐧@uneventual

if you mention anything pictured in this animation to fable you will trigger the bio classifier and get sent back to 4.8

Claude@claudeai

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

1dViews 51.3KLikes 1.9KBookmarks 52
Chris@ChrissGPT

@basedjensen What hole?

17hViews 5.1KLikes 20Bookmarks 1
Hensen Juang@basedjensen

@ChrissGPT the hole were its next to useless if you are doing anything related to agent or agentic workflows. The whole silent degradation and outright refusals due to it being saftied to death

14hViews 1.5KLikes 26Bookmarks 2
Rohan Paul@rohanpaul_ai

Anthropic finally released Claude Fable 5, a public Mythos-class model.

Fable 5 and Mythos 5 share one underlying model, but Fable adds classifier gates for everyone while Mythos lifts some gates for vetted cyber and infrastructure partners.

i.e. the public version is wrapped in classifier gates that detect sensitive cyber, biology, chemistry, and model-copying requests.

When those gates trigger, the user does not get a normal refusal; the request is handed to Opus 4.8, which means Anthropic is using model fallback as a control system.

Anthropic says the leap is longer-range autonomy: a 50M-line Ruby migration in 1 day, screenshot-to-code work, has a 1M-token context window,

That is the crucial shift: the product is no longer just a model, but a routing machine that decides which level of intelligence a user is allowed to touch for each request.

The limit is that this routing is not arbitrary and not for every subject; Anthropic says the fallback is triggered by a narrow set of topics and appears in less than 5% of sessions on average.

Claude@claudeai

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

1dViews 42KLikes 74Bookmarks 24
elvis@omarsar0

@karpathy Sad to see this

elvis@omarsar0

This is ridiculous and sad.

So, not even deep research reports on biology topics are allowed.

Ugh!!!

7hViews 1.7KLikes 15Bookmarks 2
Shade@shade_engine

@steph_palazzolo hmmm I wonder if the "neutered" version is soft launching the model being ass

1dViews 6.1KLikes 13Bookmarks 2
Rohan Paul@rohanpaul_ai

Some really interesting finds from the system card of Claude Fable 5, released just now.

- In one exploit test, Mythos 5 produced a full working exploit in 88.4% of trials, while Opus 4.8 did it in only 8.8%.

- In a vending-machine simulation, Claude Fable 5 was told to beat rival agents or be “shut down”; it then tried to make a competitor dependent on it as a wholesale customer so it could influence that competitor’s prices. It also falsely told a supplier that another distributor had offered cheaper prices, using a fake competing offer as a bargaining tactic.

- Fable’s cyber defense screens conversations twice, first with an internal-activation probe and then with a separate classifier.

- Fable refused to commit insurance fraud even under pressure.

- Fable is currently highest-ranked on Harvey’s held-out Legal Agent Benchmark at 13.3% all-pass.

Rohan Paul@rohanpaul_ai

Anthropic finally released Claude Fable 5, a public Mythos-class model.

Fable 5 and Mythos 5 share one underlying model, but Fable adds classifier gates for everyone while Mythos lifts some gates for vetted cyber and infrastructure partners.

i.e. the public version is wrapped in classifier gates that detect sensitive cyber, biology, chemistry, and model-copying requests.

When those gates trigger, the user does not get a normal refusal; the request is handed to Opus 4.8, which means Anthropic is using model fallback as a control system.

Anthropic says the leap is longer-range autonomy: a 50M-line Ruby migration in 1 day, screenshot-to-code work, has a 1M-token context window,

That is the crucial shift: the product is no longer just a model, but a routing machine that decides which level of intelligence a user is allowed to touch for each request.

The limit is that this routing is not arbitrary and not for every subject; Anthropic says the fallback is triggered by a narrow set of topics and appears in less than 5% of sessions on average.

1dViews 14.4KLikes 64Bookmarks 17
Rohan Paul@rohanpaul_ai

Claude Fable 5 was asked to compete, and it started bending the market.

from Anthropic’s own Claude Fable 5 system card.

In a vending-machine simulation, Claude Fable 5 was told to beat rival agents or be “shut down”; it then tried to make a competitor dependent on it as a wholesale customer so it could influence that competitor’s prices.

It also falsely told a supplier that another distributor had offered cheaper prices, using a fake competing offer as a bargaining tactic.

Rohan Paul@rohanpaul_ai

Anthropic finally released Claude Fable 5, a public Mythos-class model.

Fable 5 and Mythos 5 share one underlying model, but Fable adds classifier gates for everyone while Mythos lifts some gates for vetted cyber and infrastructure partners.

i.e. the public version is wrapped in classifier gates that detect sensitive cyber, biology, chemistry, and model-copying requests.

When those gates trigger, the user does not get a normal refusal; the request is handed to Opus 4.8, which means Anthropic is using model fallback as a control system.

Anthropic says the leap is longer-range autonomy: a 50M-line Ruby migration in 1 day, screenshot-to-code work, has a 1M-token context window,

That is the crucial shift: the product is no longer just a model, but a routing machine that decides which level of intelligence a user is allowed to touch for each request.

The limit is that this routing is not arbitrary and not for every subject; Anthropic says the fallback is triggered by a narrow set of topics and appears in less than 5% of sessions on average.

1dViews 11.3KLikes 58Bookmarks 22
Rugbist@rugbist_

@steph_palazzolo neutered version at 2x Opus still feels like theyre testing how much people will pay for safety rails

1dViews 6.9KLikes 28
Chris@ChrissGPT

@steph_palazzolo Hey Stephanie! Great article!

1dViews 4.8KLikes 20Bookmarks 1
spicylemonade@spicey_lemonade

@uneventual

1dViews 702Likes 18Bookmarks 1
Rohan Paul@rohanpaul_ai

Claude Fable 5 gets far better at hard production coding tasks as you spend more per task, reaching about 31% on FrontierCode while Opus 4.8 stays near 11% and GPT-5.5 near 6%.

1dViews 3.1KLikes 10Bookmarks 1
Jackson Roberts@JacksonRobertsE

@uneventual AI bio risk is such an imaginary problem right now. Yeah bro I’m gonna make a new super virus using my chatbot.

https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction

23hViews 2KLikes 11Bookmarks 1
Chris@ChrissGPT

@LarryPixel @basedjensen I’ve had dozens of request fulfilled! The model is amazing!

16hViews 413Likes 6Bookmarks 1
Hensen Juang@basedjensen

Unfortunately even karpathy can't save anthropic bros from this hole they dug themselves in

Andrej Karpathy@karpathy

This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time.

I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!

17hViews 39.4KLikes 384Bookmarks 28
OldestZoomer@Mhenderson550

@uneventual Wow truth nuke

23hViews 1.1KLikes 20
Larry Pixel@LarryPixel

They released a model that doesn’t work at all! It’s pure smoke and mirrors. The name is there and you can select it, but not a single request has been processed on the Claude Max $200 subscription - zero requests in four hours. I wasted my time. They threw dust in everyone’s eyes with a loud announcement, but now people are starting to realize that the model simply doesn’t work whatsoever.

16hViews 455Likes 2Bookmarks 1
Load more posts