Big Idea

The argument in one line.

Fable 5 is a deliberately hobbled version of the unreleased Mythos model — the safeguards that reroute 75% of queries to Sonnet are the price of giving the public access to a model Anthropic once considered too dangerous to ship.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use Claude regularly and want to know whether the new models are worth switching to or paying more for.
You're a non-technical creator or builder trying to cut through the benchmark noise and understand what actually changed.
You're curious about the Mythos/Fable naming split and why Anthropic is withholding the full model from the public.

SKIP IF…

You want a deep technical or benchmark analysis — this is an accessible overview, not a rigorous evaluation.
You're already following the Anthropic announcements closely and have read the original article.

TL;DR

The full version, fast.

Anthropic launched Claude Fable 5, a consumer-accessible version of its restricted Mythos-class model, with safety guardrails that reroute flagged queries to Sonnet 4.6. On agentic coding benchmarks it scores 80.3% versus Opus 4.8's 69.2%, and Stripe reported it compressed months of engineering into days on a 50-million-line Ruby codebase. Vision is significantly upgraded — it can reconstruct a web app from a screenshot alone. Pricing is $10/million input and $30/million output tokens, exactly double Opus 4.8. The full Mythos 5 model remains restricted to government and vetted cyber-defense organizations.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:03

01 · What Was Just Released

Fable 5 is a Mythos-class model made safe for general use — a consumer version of the model Anthropic previously said was too dangerous to ship.

01:03 – 02:44

02 · The Safety Reroute Problem

About 75% of real-world prompts triggered the safety guardrail during testing, dropping users to Sonnet 4.6 instead of Fable 5 — a significant usability friction point.

02:44 – 04:01

03 · Mythos 5 and Project Glasswing

The full Mythos 5 model — same underlying weights as Fable 5 but with safety lifted — is being deployed through the US government's Project Glasswing for cyber defense.

04:01 – 05:10

04 · Benchmarks vs. Opus 4.8

Fable 5 scores 80.3% on agentic coding benchmarks versus Opus 4.8's 69.2% — an 11-point jump. GPT-5.5 and Codex no longer hold the coding crown according to these numbers.

05:10 – 07:03

05 · Software Engineering and Vision

Stripe reported months of engineering compressed into days. Fable 5 leads frontier models even at medium effort, and its vision capability can rebuild a web app from a screenshot.

07:03 – 08:29

06 · Pricing Breakdown

$10/million input and $30/million output — exactly double Opus 4.8, far less than the 10x premium many predicted. Developers can access it through the Claude API.

08:29 – 09:30

07 · Drug Design and Recommendation

Mythos 5 accelerated protein design by 10x and matched human scientists with no human assistance. Final verdict: stick with Opus 4.8 unless you're a power user, given the double price and reroute friction.

Atomic Insights

Lines worth screenshotting.

Fable 5 is described by Anthropic as a watered-down Mythos — the same underlying model with safety guardrails that block cybersecurity-adjacent queries.
About 75% of real-world prompts triggered the safety reroute during pre-release testing, redirecting to Sonnet 4.6 instead of Fable 5.
Agentic coding benchmark: Fable 5 scores 80.3% versus Opus 4.8's 69.2% — an 11-point jump in one model generation.
Stripe reported Fable 5 compressed months of engineering work into days on a 50-million-line Ruby codebase.
Fable 5 is the first frontier model claimed to beat benchmarks even at medium effort level, making it cost-effective for lighter workloads.
Vision upgrade: Fable 5 can extract precise numbers from scientific figures and rebuild a web app's source code from screenshots alone.
Mythos 5 initially deploys through Project Glasswing in collaboration with the US government — the same pathway used for Mythos preview.
Mythos-class models excel at discovering and exploiting software vulnerabilities, making cybersecurity the primary reason for restricted access.
Fable 5 pricing is $10/million input and $30/million output tokens — exactly double Opus 4.8, not the 10x premium some predicted.
In drug design, Mythos 5 accelerated protein design by 10x and matched skilled human scientists with no human assistance.
9 of 14 protein targets from Anthropic's internal drug design study yielded strong candidates for drug design currently under investigation.
The Fable/Mythos naming architecture signals a two-tier public/restricted release strategy Anthropic plans to repeat for future capability jumps.

Takeaway

What Fable 5's safety tax means for your workflow

WHAT TO LEARN

The gap between what Fable 5 can do and what it will do for most prompts is the most important thing to understand before upgrading.

01What Was Just Released

Fable 5 is not a new architecture — it is Mythos 5 with safety guardrails applied, making it a constrained version of a model Anthropic previously withheld entirely.

02The Safety Reroute Problem

Fable 5's safety reroute triggered on roughly 75% of real-world prompts during pre-release testing, even on non-sensitive content — a significant practical limitation.
When rerouted, users get Sonnet 4.6, not Fable 5 — the quality drop is real and the reroute is silent except for a 'chat paused' notice.

03Mythos 5 and Project Glasswing

Mythos 5 is the same model as Fable 5 with some safety restrictions lifted — it is only available through the US government's Project Glasswing and to vetted cyber-defense organizations.
Anthropic's rationale for the two-tier release: Mythos-class models can discover and exploit software vulnerabilities at scale, making unrestricted public access a genuine risk.

04Benchmarks vs. Opus 4.8

An 11-point benchmark improvement in agentic coding (69.2% to 80.3%) in one model generation is a meaningful jump — the question is whether real-world tasks match benchmark conditions.
Benchmark numbers from model providers are self-reported and should be treated as directional, not absolute, until independent third-party testing confirms them.

05Software Engineering and Vision

Stripe's 'months into days' claim on a 50-million-line Ruby codebase is the strongest real-world validation in the announcement — and it comes from an enterprise with actual production scale.
The vision upgrade (rebuilding web apps from screenshots) matters most for computer-use workflows where the model needs to interpret what's on screen accurately.

06Pricing Breakdown

At double Opus 4.8's price, Fable 5 is only worth it if you're running long agentic tasks where the 11-point quality gain translates to fewer retries and less manual correction.
The effort level dial (low/medium/high/extra/max) gives you a lever to trade quality for cost — medium effort Fable 5 may outperform Opus 4.8 at high effort for less money.

07Drug Design and Recommendation

Mythos 5 matching human scientists in protein design with no human assistance is a category boundary event — not a tool assisting a researcher, but a system replacing one.
For most users, Opus 4.8 remains the better daily driver: half the price, no safety reroute friction, and adequate for the vast majority of creative and coding tasks.

Glossary

Terms worth knowing.

Mythos-class model: Anthropic's internal classification for its highest-capability models deemed too powerful or risky for general public release. Mythos models are made available only to vetted organizations, governments, or cyber-defense programs.
Fable 5: Anthropic's consumer-accessible version of its Mythos 5 model, released with safety guardrails that reroute queries on high-risk topics to a less capable model.
Project Glasswing: A US government initiative through which Anthropic is deploying Mythos 5 to vetted cyber-defense and infrastructure organizations, building on the earlier Mythos preview partnership.
Agentic coding: Tasks where an AI model autonomously writes, executes, debugs, and iterates on code across a large codebase with minimal human intervention — closer to a software engineer than a code autocomplete tool.
Effort level: A user-selectable parameter in Claude's interface (low, medium, high, extra, max) that controls how much computational work the model does on a response. Higher effort costs more tokens.
Input/output tokens: The units Claude bills by. Input tokens are the text you send; output tokens are what the model writes back. Pricing is set per million tokens of each type.

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

00:00Anthropic just dropped Claude Fable five and Claude Mithos five. So I read through this entire article and have been playing around with it myself. As you could see, I have this directly inside of Claude now.

00:10So in this video, I'm gonna break down this entire article so you don't have to read the whole thing. And do not worry. I'm not gonna sit and just hype this up.

00:17I'm gonna break down key things that change that we need to note. We're gonna talk about availability, pricing, what matters, and what we could actually expect from this new model release from Anthropic.

00:28So without further ado, let's dive right into this. Alright. So first things first, if you do wanna read this entire article, it is pretty hefty.

00:34There's gonna be a link in the description for you to do so. However, I went through and made a bite sized breakdown so you could understand the TLDR of this entire release.

00:42I don't wanna waste any more time. Let's dive right into these talking points that I found after reading this article. So Anthropic said, today, we're launching Claude Fable five, a mythos class model we've made safe for general use.

00:54So real quickly, if you don't know what Claude mythos is, this is the model that Anthropic gave to some of the biggest companies a couple of months ago in order to test out the capabilities because they did state that this model was too powerful to release to the public, and there was tons of potential for cybersecurity issues.

01:10So this is a watered down version of that mythos model that they're now releasing to the consumer public like you and I. Now it says here, Fable five's capabilities exceed those of any model we've made generally available. It is state of the art on nearly all test benchmarks of AI capability, vision, scientific research, and many other areas.

01:29The longer and more complex tasks, the larger Fable five's lead over other models. Now let's quickly talk about the risks before we break down some of the benchmarks and talk about actually how good this model is. So Anthropics stated that releasing a model with this capability comes with a risk.

01:44So without safeguards, Fable five's capabilities in areas like cybersecurity could be misused and cause serious damage. We've therefore launched the model with safeguards that means queries on some topics will instead receive a response from our next most capable model, Claude Opus 4.8.

01:59And I do wanna talk about this real quickly because in preparation for this video, I've been testing out the Fable five model, and I would say about 75 of the prompts that I gave it, I would get this error message right here. It says chat paused.

02:13This model has safety measures that flags messages on most cybersecurity or biological topics. They may flag safe normal content as well, and it basically prompted me to continue with this prompt using Claude Sonnet 4.6 or other models.

02:27So this means if there's something that Anthropic maybe flags that could be high risk to use this model for, it's gonna route you to the next best model that they think will be good to actually answer that question, which, to be honest, after testing this a little bit, you know, before this video, has been quite annoying.

02:42You know, like I said, probably 75% of the time, I've prompted Fable five and tried to get a response to see what it's capable of. I get that message saying it needs to use a different model, which is pretty annoying.

02:52Now moving on to the next point here, it says that they are releasing to a small group of cyber defenders and infrastructure providers. They are launching Claude Mythos five. And a key point here is it says this is the same underlying model as Fable five, but with the safeguards lifted in some areas.

03:08Mythos five will initially be deployed through Project Glasswing in collaboration with the US government as an upgrade to Claude Mito's preview. Very interesting as they're giving the US government access this to make sure that, you know, this isn't being used by malicious people, which I guess makes sense. I don't wanna get political here, but it sounds like Claude Mito's five is a next step above Fable five, and they don't wanna just release this to the public yet.

03:31And that's been the case over the past couple months since they have been doing this prior. But it sounds like now they're releasing Claude Mithos five to, you know, other people as well. Then it says, in consultation with the US government, we plan to steadily expand access to clogged methos five, just like I stated above.

03:45And, again, if you don't have any prior context on methos, the model that, you know, has been out for a couple of months that they released to only, you know, a certain amount of companies, cybersecurity is the main issue here. So methos class models excel at discovering and exploiting software vulnerabilities. They can make cybersecurity attacks substantially easier and cheaper to commit.

04:05So those are some of the risks we already knew that Anthropic was trying to handpick who they were gonna give access to Claude and Mythos to. Now let's talk about the benchmarks real quickly with Fable five and Mythos five. So I don't wanna speak in only a developer language here, so let's quickly just talk about this and what actually matters.

04:21Now Opus 4.8 is the model that they just released literally about a week and a half ago. And as you could see for Agentic coding, that was 69.2%, whereas Claude Mithos five and Fable five are at 80.3 for Agentic coding.

04:38And this, to me, honestly, is a huge step up. So I'm really interested to see this in practice when using something like ClaudeCode because everybody and their mom is talking about GPT 5.5 and Codex right now saying it's the best coding model, but, apparently, according to this benchmark, that is not the case anymore.

04:55Then again, I don't wanna bore you going through all of these numbers because to be honest, sometimes I feel like this can be a bit fabricated and, honestly, it doesn't really mean much until we test it out and, you know, try it ourselves. But all we need to know is this is basically better than all these models aside from Claude Mito's preview, which is the model that, you know, isn't released to the general public yet.

05:15Now here's the part of the article where it talks about software engineering. So it said during early testing, Stripe reported that Fable five compressed months of engineering into days, which is absolutely crazy if you think about it.

05:29The fact that it used to take real life software engineers months to write this code, it can now do that in days with this new model. And, again, this is with the Fable five model. This isn't even Mythos five in Mythos five preview, which, you know, the public doesn't have access to yet.

05:44Then if I come down here, it says Fable five scores highest among frontier models even at medium effort level. So to show you what I mean, if I come back over to Claude, I could actually go ahead, click on, you know, this little button here, and we could see our different models that we could choose from. We have Fable five that says included until June 22.

06:02We have Opus 4.8, which again dropped to, like, about a week, a week and a half ago. And then we could also see we have this effort level here that we could select. So it's always gonna default to high.

06:11However, we have extra and we have max as well as medium and low if we ever wanna use those. But what Anthropic is saying is that even if we are on the medium level, this is better than all of the other Frontier models out there, which is good news if you wanna use a better, more capable model but not have to pay the actual pricing for Fable five high.

06:30And stick with me because we're gonna talk about pricing later on in this video, and this really did surprise me in terms of how much it costs. Next, apparently, there's a pretty big upgrade when it comes to vision. So Fable five is a new state of the art model for tasks including vision.

06:43It can extract precise numbers from detailed scientific figures and can perform complex vision based tasks like rebuilding a web app source code from screenshots alone, which is pretty crazy. And, again, this goes to show that it's not only good at vision tasks, but it's also really good when it comes to agentic coding as well.

07:01And vision is one of the big things important for these models that I think we really need to master. Because when it comes to being able to use computer use where Claude and these different models can go and use our computer, it needs to be really good at being able to see what's actually on our screen.

07:17Now on to the part that everybody is probably interested in. This is pricing. Pricing for both of these different models, both Mythos five and Fable five, these are gonna cost $10 per million input output tokens.

07:33And you might not be familiar with what that even means, so let's break this down a little bit more. Let's compare this to OPUS 4.8, which is the model that just came out a couple of weeks ago. This, right now, if you're using OPUS 4.8, is 5,000,000 input tokens and 25,000,000 output tokens.

07:48So this means that both Fable and Mythos are gonna be two times more expensive than the Opus models, which, to be honest, isn't as much as people were assuming. Some people on Twitter were thinking that this was gonna be maybe 10 times more expensive.

08:04So I guess this is good news for us if we do wanna use these models and not burn through all of our tokens inside of our clawed accounts. Now this is another thing that's pretty interesting. So it says, drug design.

08:14Using Methos five, our internal protein design experts accelerated aspects of the drug design process by around 10 times. And then this part right here is pretty mind blowing, so stick with me. It says, in one example, they found that Methos five with protein design and bioinformatics tools, but no human assistance matches or beats skilled human operators.

08:33In doing so, the model executes all of the tasks that are normally completed by a scientist. I know nothing when it comes to medical research and drug design, but if Anthropic is claiming this, this is pretty crazy when we do think about it. So this should be available to everybody now.

08:48I at least believe this will be available to the pro plans and the max plans, but I'm not sure if it's gonna roll out instantly to the free plan. But go ahead, check inside of either your Claude desktop app or your Claude web app, and then you should be able to see this mode right here where it says Fable five high. You could select this.

09:04You could always choose from Opus 4.8, which, honestly, if you are using Claude and you're not a power user, I would probably suggest not to use Fable five right now because it is double the price, and I know a lot of you guys are already running into issues when it comes to token usage inside of your Claude account. I'll be testing this ruthlessly over the coming days, so make sure to subscribe for more videos like this if you are interested.

09:26And, anyways, I hope you guys have a good video, and I look forward to seeing you in the next one. Cheers.

The Hook

The bait, then the rug-pull.

Anthropic published its most consequential model announcement in months, and the real story isn't the benchmark numbers — it's what they're still not letting you use. Brock Mesarich reads the full release article live, annotates the key passages, and stress-tests Fable 5 before the video ends.