AI - Anthropic release Claude Fable 5.2 and Mythos 5.2, faster and cheaper

Anthropic has done it again, and this time the headline is not only “more capable” but also “faster AND cheaper” 🤣

Two months after Fable 5.1 and Mythos 5.1, we now have Claude Fable 5.2 and Claude Mythos 5.2. A price cut, a speed bump and a pile of slightly mind-bending demos from Anthropic… and, of course, the internet immediately split into “this is insane” and “I smell a catch” 😉

What I am personally most happy about is the free Fable reset, as that will definitely help with testing it 😊

Claude Fable 5.2 and Mythos 5.2

As with the 5.1 release, Fable and Mythos are the same underlying model with different safeguards. Fable 5.2 is available to everyone, with classifiers that block high-risk cybersecurity and biology tasks. Mythos 5.2 drops some of those safeguards and is only available to vetted organisations through Anthropic’s Cyber and Life Sciences Verification Programs.

Anthropic positions both models above Opus 5.5 for the most ambitious, open-ended technical work. For most everyday coding and knowledge work, Anthropic still recommends starting with Opus 5.5 and moving up to Fable only when you need it.

Official announcement

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

Introducing Claude Fable 5.2 and Claude Mythos 5.2

Claude Fable 5.2 and Claude Mythos 5.2 show improvements in areas like coding, reasoning, and vision. During prerelease testing they’ve helped our engineers complete otherwise-infeasible software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made substantial progress on robotics evaluations, and formalized the proof of Fermat’s Last Theorem in 30 hours—among many other applications we’ll discuss below.

What is new?

According to Anthropic:

• Faster: outputs are generated up to 30% faster than Fable 5.1 and Mythos 5.1.
• Cheaper: $8 per million input tokens and $40 per million output tokens, down from $10 and $50.
• Optional fast mode: up to 2.5 times faster token generation for $16 per million input tokens and $80 per million output tokens. Mythos fast mode is only available to organisations in the Verification Programs.
• More efficient: on Medium effort (the default in the Claude apps and Claude Code), Fable 5.2 can complete challenging tasks with fewer tokens than Fable 5.1. Anthropic says that, in early testing, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort on FrontierCode 2.0, for roughly half the cost per task.
• Fewer unnecessary blocks: Anthropic says it saw about 55% fewer safety classifier blocks with Fable 5.2 than with Fable 5.1. When a query is blocked, you get a response from Opus 4.8 (cybersecurity) or Opus 5 (biology) instead.
• Advisor and Orchestrator roles: Fable 5.2 can be consulted by Haiku, Sonnet or Opus on harder problems, or can delegate simpler tasks to smaller models.
• Model ID: claude-fable-5-2, with a 1M token context window and 128K maximum output. The API defaults to high effort, and adaptive thinking is always on.
• Availability: all paid Claude plans, the Claude Platform, Claude Code, Amazon Bedrock, Google Cloud and Microsoft Foundry.

Anthropic also says that it deliberately limited the model’s progress in a few AI research areas during training, such as accelerator kernel optimisation and improving large language model architectures, so in those areas Fable 5.2 performs at about the level of Opus 5.5. Remember that one, because it comes up again later 😉

Pricing and speed

Fable 5.2 is now available in the Claude Platform and Claude Code.

It’s up to 30% faster than Fable 5.1, and costs $8/$40 per 1M tokens, down from $10/$50.

Some people were, let us say, rather excited about it 🤣

ANTHROPIC, WHAT THE ACTUAL FUCK???

Claude Fable 5.2 isn’t just more capable. It’s also faster than its predecessor.

• $8/M input tokens, down from $10
• $40/M output tokens, down from $50
• 20% cheaper
• Up to 30% faster

WHAT IS GOING ON AT ANTHROPIC??

Omg wtf?? Claude Fable 5.2 is an absolute monster of a model.. mogging Opus 5.5 on literally everything..

And its 20% cheaper: $8/M input and $40/M output, down from $10/M and $50/M for 5.1..

My bad that I slept and missed the launch..

It is a 20% cut per token, but it is still twice the price of Opus 5.5 per token. A cheaper token does not automatically mean a cheaper job either; the total cost depends on how many tokens the model actually burns, which effort level you use and how long your agents keep running. We will see some real-world examples of that further down 😉

Free Fable resets and a week of Fable on Pro

This is the bit I am super happy about 😊 To mark the launch, Anthropic is giving eligible paid subscribers extra Fable usage.

Loyalty pays off.

Three Fable resets for subscribers who’ve been with Claude Code since the beginning.

The ClaudeDevs post that he is quoting has the breakdown:

We’re giving Ultra and Team Premium users Fable resets based on how long you’ve subscribed:

<6 mo: 1
6–12 mo: 2
>12 mo: 3

Use them anytime by March 8.

Pro and Team Standard get a week to try Fable 5.2, up to half your weekly limit, plus one reset.

ICYMI 🔥: Claude Fable 5.2 is included for up to 50% of Claude subscription until November 16.

Testing allowance 👀

“Fable is included for up to 50% of your plan limits until November 16. After that, it’s available with usage credits.”

🚨 Anthropic is giving Claude users FREE Fable usage resets!

Users receive 1–3 free resets depending on subscription history

I got 1 reset and it expires 17 November

Anthropic is making its most powerful models more accessible than ever

Fable 5.2 in GitHub Copilot and Microsoft Foundry

Fable 5.2 is also available outside Anthropic’s own apps, including GitHub Copilot and Microsoft Foundry.

Claude Fable 5.2, Anthropic’s newest Fable model, is now generally available in GitHub Copilot and Microsoft Foundry.

In early testing, Fable 5.2 solved more tasks than 5.1 with fewer steps and tool calls.

According to the GitHub changelog, Fable 5.2 is available to Copilot Pro+, Max, Business and Enterprise users in VSCode, Visual Studio, Copilot CLI, the Copilot coding agent, the GitHub Copilot app, github.com, GitHub Mobile, JetBrains IDEs, Xcode and Eclipse. It is billed at provider list pricing under usage-based billing.

What Anthropic says it built with it

This is where it gets a bit wild. These are Anthropic’s claims from prerelease testing, not independent results, but they are certainly ambitious 🤣

A team of 1,000 Fable 5.2 agents fully formalized Fermat’s Last Theorem in 30 hours, translating the proof into a language a computer can check step by step.

With the same setup, Opus 5.5 completed only 26% in 18 hours, and barely made any more progress after 30 hours.

This is INSANE. Anthropic says one of its engineers asked Claude Fable 5.2 to build an operating system from an empty folder.

The model coordinated over 1,000 subagents working around the clock for eight days and produced a working prototype OS that boots on a laptop, complete with disk, keyboard and network drivers, file storage, windows, a login screen, an installer and built-in Claude functionality.

This is INCREDIBLE.

You can now ask Fable to make any smart home device/gadget you want!

The Claude post he is quoting explains:

From a one-paragraph prompt, Fable 5.2 designed three working electronic devices, including this home voice assistant hub.

According to Anthropic, the three devices were a home voice-assistant hub, a USB speakerphone and a handheld virtual pet, built with Diode, a company that manufactures circuit boards. All three worked without electrical modifications. “Any smart home device you want” is a bit of a stretch, but it is still very impressive 😉

Anthropic says Fable 5.2 just did some things that are honestly hard to wrap your head around:

→ Solved two open problems in black hole physics.
→ Formalized the proof of Fermat’s Last Theorem in just 30 hours.
→ Coordinated over 1,000 AI agents to build a working operating system from scratch.

Think about that for a second.

We’re talking advanced physics, one of mathematics’ most famous problems, and an entire OS built through AI collaboration.

If these claims hold up, this is way beyond another model getting smarter.

Are we actually getting closer to AGI, or is it still too early to call it? 😭

“If these claims hold up” is the key part. The physics results are published as preprints, which is a good start, but I would like to see what the wider scientific community makes of them before getting too carried away 😊

Benchmarks

🚨 Fable 5.2 is looking seriously impressive.

It beats Opus 5.5, GPT-6 Astra and Gemini 4 Carbon on 13 of 14 benchmarks in this comparison.

Some standout results:

• 89.6%: CVE Rediscovery
• 89.3%: Computer Use
• 85.1%: Protein Design
• 82.1%: Agentic Search
• 76.3%: Cybersecurity
• 74.6%: Scientific Research
• 74.2%: Humanity’s Last Exam

It’s not just coding.

Fable 5.2 is showing strong performance across cybersecurity, robotics, scientific research, computer use and knowledge work.

Fable 5.2 benchmarks:

These numbers come from Anthropic’s own benchmark table. A few of them (such as CVE Rediscovery and the cybersecurity score) are for Mythos 5.2 rather than Fable 5.2, and Anthropic notes that Fable’s safeguards block most cybersecurity tasks. Even in Anthropic’s table, Fable 5.2 does not win every row: Opus 5.5 and Gemini 4 Carbon each come out slightly ahead on one.

Fable 5.2 scored 65 on the AA Intelligence Index.

That puts it at the top of the current model rankings, making it the strongest model available by this measure.

Anthropic has raised the bar once again.

OpenAI, your move.

This is from an independent index rather than Anthropic, which is nice to see, but it is still one index 😉

Tips from the people who built it

Some of the most useful posts come from people at Anthropic who have been using it internally. Interestingly, they are not saying “use Fable for everything”.

I really love using Fable 5.2. On big, open-ended work it feels like a major step above Opus 5.5.

A few tips for getting the most out of it:

• Ask what’s wrong with your idea before it builds anything. It’s great at brainstorming and will push back
• Start on med/high effort, and save xhigh for hardest problems
• Tell it up front what needs your approval (e.g. deleting things) and that the rest is its call
• Give it a whole project instead of a next step and let it run for hours. This is where it shines

We’re releasing Claude Fable & Mythos 5.2! They’re good models, I have some demos and tips for you.

Fable beats Opus on big, long tasks. Good examples are rewriting an entire codebase, creating an excellent first draft of a large artifact, or exploring ambiguous scenarios.

I expect to see it used to port a lot of stuff to Rust, support journalists in large-scale investigations, and help with research & science. I often still prefer Opus 5.5 for everyday work. Fable’s got a big brain and thinks a little slower.

Claude Fable 5.2 and Mythos 5.2 are out today!

They’re super sharp to work with. On complex problems you can really feel the extra intelligence; researchers working with Fable 5.2 solved two problems from theoretical physics.

Fable is also a great thought partner on strategy, paired with Opus 5.5 for a lot of my work.

Can’t wait to see how you use Fable 5.2 to bring your big ideas to life!

from using Mythos/Fable 5.2 internally, this is one of those “big model smell” models — has a je ne sais quoi at writing, 3d modeling, making videos, 1-shotting interactive tools, and robotics. Also the best for cyberdefense.

The common theme is to use Fable for big, long, open-ended work, and to keep Opus 5.5 as the daily driver. That matches Anthropic’s own guidance, so it seems to be the consistent message 😊

The prompting guide

Anthropic has also published a Fable 5.2 prompting guide, which is worth a read before moving your existing prompts and agents across.

Anthropic’s Fable 5.2 prompting guide has some interesting findings.

• High effort is the sweet spot. Max isn’t always better.
• Agents may stop before finishing tasks, especially at lower effort.
• Outdated instructions can make agents overly cautious.
• Giving agents code ownership improves refactoring but increases cost.
• Avoid unnecessary subagents, especially at higher effort.
• Telling agents they have a large token allowance can increase costs.
• Telling agents not to revisit previous answers can make them miss their own mistakes.

I put together a simple chart showing which model and effort level makes sense for different tasks.

Anthropic just released the official Fable 5.2 prompting playbook

It’s f*cking detailed…

→ calibrate effort + finish every task + ownership language + progress notes

→ subagent limits + caution cleanup + migration checks + readable handoffs

copy-paste prompts that stop agents from quitting or rewriting everything, then read the full “Context, Harness, Loops, Prompts” guide ↓

And then there are the inevitable “one prompt to rule them all” posts 🤣

Fable 5.2 is insane for building a team of coding agents…

send this prompt to Claude Fable 5.2 and it will turn into a team of 20+ senior developers at 10% of the cost

then read the full guide to building self-improving agent team Fable in the article below ↓

I would take “20+ senior developers at 10% of the cost” with a very large pinch of salt. Anthropic’s own guide warns against unnecessary subagents because they can account for a large part of what a session costs, so a giant agent team is not automatically a cheap one 😉

Head-to-head tests

Claude Fable 5.2 vs Claude Opus 5.5 🧊

The first Fable 5.2 result is here. And it’s totally insane. 🤯

Same prompt. Cathedral of Ice.
Fable 5.2: $81.10 · 1h 56m 52s
Opus 5.5: $45.97 · 3h 11m 26s

I know I say it every time, but this is a huge leap. A huge jump. 😱

Fable cost more, but it finished 1h 15m faster. And just look at the ice, the ship, the whale. 👀

The future looks bright and amazing. 🔥

What do you guys think?

This is a nice illustration of the cost point. Fable finished more than an hour faster, but the run cost almost 80% more than Opus 5.5 did. Whether that is worth it really depends on whether your time or your token bill matters more 🤣

Claude Fable 5.2 High vs Opus 5.5 High.

One buggy Python function. 8 bugs planted. Same prompt, fresh chats, first answer only.

Both full answers, the answer key and the code are here:

https://www.asumacodes.xyz/arena/bug-hunt?utm_source=x&utm_medium=social&utm_campaign=fable-vs-opus

Find them yourself before you scroll 👇

I like this one because the answers and the answer key are published, so you can judge for yourself rather than trusting a screenshot 😊

The more sceptical view

Not everyone is impressed, and I think these are worth reading alongside the hype.

My First Impressions of Fable 5.2

I’ve been using it for a few hours now, and I have a few thoughts on the new Fable 5.2:

1. The weirdest RL I’ve ever seen. On some tasks it’s a genius; on others, it’s as dumb as a shrimp. In pure coding tasks, it’s at least on par with Opus 5.5. But ask it to reason through something non-standard that isn’t coding-related, and its output drops to the level of GPT-3.5. I asked both Fable 5.5 and GPT-6 Astra to flesh out a project idea and Fable’s response was complete garbage

2. Fable models have always been notorious for mediocre taste - worse than Opus. This time, Fable surprised me with a complete absence of it. In a long list, it inexplicably added two separate pagination controls: one expands the currently loaded list of data, and the second one underneath fetches new content. On top of that, you can see plenty of demos on X with pretty mediocre results that are even worse than Opus

3. It seems to be the most token-hungry model in the world. The advertised price cut makes zero sense when Fable 5.2 burns through four times more tokens than 5.1

4. The most pedantic model yet. It constantly tries to prove you wrong. Tell it that snow is white, and it will find a million reasons why that’s not entirely accurate. GPT-6 Astra had this annoying trait as well

5. It’s noticeably faster. On medium effort, it responds almost instantaneously

6. Fable 5.5 has picked up a trait that used to belong to Sonnet 5.5 - diligence. Once it takes on a task, it will check and iterate multiple times until it achieves a solid result

7. For coding tasks, there is zero difference compared to Opus 5.5. Across all other aspects, Opus 5.5 is, in my subjective view, superior to Fable 5.5

I’ve never really liked Fable models. There’s not much point to them. They’re stubborn, weighed down by guardrail filters, have poor taste, and are expensive, token-hungry, and in some areas worse than the Opus lineup. This one is no exception. In my view, the new Fable 5.2 makes no sense, and I think I understand why they called this release 5.2 rather than 5.5. A revolutionary leap like the jump from Opus 5 to 5.5 just didn’t happen

(I assume the “Fable 5.5” mentions are meant to be Fable 5.2 😉)

The “four times more tokens than 5.1” claim directly contradicts Anthropic’s statement that Fable 5.2 uses fewer tokens than 5.1 on Medium effort. It is one person’s experience after a few hours, but it is exactly the kind of thing that decides whether the price cut is real for you. The “pushes back a lot” point also lines up with Alex Albert’s tip that it “will push back”, so whether that is pedantic or useful probably depends on the day 🤣

I smell a catch with Fable 5.2. There’s a reason they called it 5.2 and slashed the prices, even though everyone was expecting at least 5.5 or even Fable 6

Something’s off here, but I can’t quite figure out what yet

so I guess now we know what is the big secret Anthropic discovered and why they think “RSI is imminent”

they gimped Fable/Mythos 5.2 during training

there’s one or two areas where it’s not allowed to be better than Opus 5.5

handing out ungimped models above Opus 5.5 might be equivalent to giving out a self replicating frontier model…

As expected, Anthropic has a significantly better model ready to deploy internally, one they aren’t sharing with us peasants.

We have to make do with Fable 5.2, as they themselves indicate in their own paper.

This is the training limitation I mentioned earlier. Anthropic openly says it limited the model’s progress in specific AI research areas, such as accelerator kernel optimisation and improving model architectures, so that it performs at about the Opus 5.5 level there. Anthropic also says this did not affect agentic coding, hardware development or traditional machine learning research. The “RSI is imminent” and “secret better model” interpretations are Wes’s and Chubby’s readings of that, not something Anthropic has said, but it is fair to say it is an unusual thing to openly admit 😉

My take

On paper, this is a very nice release. Faster, cheaper per token, fewer annoying classifier blocks, and some genuinely jaw-dropping demos, even if they are Anthropic’s own demos 🤣

But the sceptics raise valid points. Fable is still twice the price of Opus 5.5 per token, the early real-world cost examples are mixed, and even the people at Anthropic are saying that Opus 5.5 is still their daily driver. I think the honest summary is that Fable 5.2 is the “big, long, hard problem” model, not the “use it for everything” model.

I have not had a chance to try it properly yet, but thanks to that free reset I will definitely be giving it a go on something big this week, and we will see whether it is a genius or “as dumb as a shrimp” 🤣

Useful links

• Introducing Claude Fable 5.2 and Claude Mythos 5.2
• Claude Fable 5.2 and Mythos 5.2 System Card
• Claude Fable 5.2 model documentation
• What is new in Claude Fable 5.2
• Claude Fable 5.2 migration guide
• Prompting Claude Fable 5.2
• Claude API pricing
• Claude Fable 5.2 launch offers
• Claude Fable 5.2 in GitHub Copilot
• Claude in Microsoft Foundry
• Formalizing Fermat’s Last Theorem with Fable 5.2
• Claude designing electronics
• Cyber Verification Program
• Life Sciences Verification Program