Xin Zhiyuan reports
Just moments ago, Claude Fable 5.2 and Claude Mythos 5.2 made a stunning debut!
In benchmark tests, it surpassed nearly every model (basically, all of them surpassed Opus 5.5!), leading across the board in intelligent coding, scientific research, cybersecurity, and computer use.
These two titans have become the “super scientist” and “chief engineer” capable of leading cutting-edge scientific research and commanding thousands of AI agents.
AI accomplished a major feat in the history of human mathematics—
In just 30 hours, it independently and completely proved Fermat’s Last Theorem for the first time,and also completed formal verification!
Fable 5.2 also solved two unresolved problems in physics involving black holes and quantum gravity.
In addition, they built an operating system from scratch, independently designed and manufactured three electronic devices, and predicted the structure of a complex adenovirus capsid.
The best part? The $20 Pro plan is now available! But the offer only lasts until November 16.
“Father of Claude Code” Boris Cherny also officially announced some good news:
The longer you stay subscribed, the more Fable resets you get, with up to three chances to try again.
After all this waiting, Anthropic has finally dropped a bombshell.
The Fable and Mythos Twins: The Ultimate Trump Card for an Epic IPO
A new king is crowned! Claude Fable 5.2 took the top spot on the AA Intelligence Index v5, scoring 65 points and ranking first overall!
Fable 5.2 also broke a record, soaring to 81% on the hardest intelligence test, ARC-AGI 3.
Fable 5.2 and Mythos 5.2 are two personas of the same underlying model. Their purpose is to tackle “the most daunting technical challenge.”
Fable 5.2 is a super brain for the masses,with exceptional coding, reasoning, and visual capabilities, but it has been fitted with safety shackles.Fable 5.2 has cured its “cleanliness OCD,” reducing the false-positive safety blocking rate by about 55%.
Today, it is fully available!
And Mythos 5.2 is the beast unleashed for geeks and elite institutions.It has no biosecurity or cybersecurity restrictions, making it the strongest “hacking master” and “biologist” on Earth.
In network vulnerability discovery (CVE Rediscovery), Mythos 5.2 found 10.2% more vulnerabilities than the previous strongest model, “with no exaggeration whatsoever.”
The most alarming thing is that their capabilities have doubled, they are 30% faster, and they’re even cheaper! ($8 per million input tokens, $40 per million output tokens).
Anthropic even launched “Fast Mode,” with speed soaring by 2.5×!
Official recommendation: For day-to-day use, choose Opus 5.5 (large cup) or Sonnet 5.5 (medium cup); only when facing an epic-scale problem is it worth summoning Fable or Mythos.
After testing it, leading quantitative trading firm Jump Trading praised it as expert-level work; this speed and autonomous judgment directly eliminated the tedious human time spent on model fine-tuning and task allocation.
In the past, building an executive and R&D team required you to spend a year and a half recruiting and offering annual salaries of one million yuan, but now it takes only 30 hours, plus computing costs of a few cups of coffee.
The explosion in productivity is no longer just a slogan; real money is now literally sitting on everyone’s desk.
The All-Star Agent Team: Proving Fermat’s Last Theorem in 30 Hours
The most astonishing part of the official release was undoubtedly Fable 5.2’s formal proof of Fermat’s Last Theorem.
Previously, the unreleased version of Claude had taken as long as 11 days to complete the proof under human guidance, an extraordinarily monumental and cumbersome process.
This time, the human researchers boldly decided to step back completely!
They gave 1,000 Fable 5.2 agents the most basic prompt, without providing any mathematical reference materials.
As a result, this all-star team of agents came up with its own method for coordinating their work and completed the formalization of the proof of Fermat’s Last Theorem in just 30 hours!
Even the generated proof code was seven times shorter than before.
By contrast, Opus 5.5 ran for 18 hours and completed only 26%, before completely freezing. This is the terrifying generational gap brought by Fable 5.2.
Insane coding: Writing an operating system in 8 days
Fable 5.2 demonstrated overwhelming coding dominance.
An Anthropic engineer ran an experiment: gave Fable 5.2 an empty folder and asked it to design a prototype operating system.
As a result, the model coordinated more than 1,000 subagents, coding, testing, and reviewing around the clock for eight straight days.
After eight days, an operating system that could actually boot on a laptop was born!
It featured a kernel to manage the machine, disk/keyboard/network drivers, persistent file storage, a windowed interface, a login screen, and even built-in Claude functionality. This is the foundation for building a computer from scratch.
The official blog revealed: Currently, more than 90% of the merged code in Anthropic's core codebase has been written independently by Claude under human guidance!
Fable 5.2 even rewrote an entire key production service from Python to Rust, boosting startup speed by 26-fold!
Vercel engineer Joe Haddad remarked:
“This model practically lives in my head. It completely understands the true intent behind my prompts. I didn’t even have to tell it once, ‘No, what I mean is…’ We had put off a project for weeks using other models, but Fable 5.2 wrapped it up in days, and human reviewers simply couldn’t find a single flaw in the code it wrote.”
The physics world’s “greatest mind”: tackling black hole open problems head-on
When Fable 5.2 stepped in, it directly solved two open problems in theoretical physics, pointing toward the ultimate holy grail of modern physics—unifying relativity and quantum mechanics.
In one problem involving “holographic duality,” physicists had set out a computational framework in 2003, but because the calculations were so immense and cumbersome, no one had completed them for more than 20 years.
Claude Fable 5.2 stepped in. In a highly autonomous fashion, it completed the arduous derivation and proved that even in the tiniest universe, where only a quantum description applies, black hole formation is as sudden as boiling water.
Moreover, it precisely identified the limits of a technique physicists had used for 15 years. Both discoveries have now been published as preprint papers.
The hardware engineers’ “grandmaster”
With no ready-made drawings, just a one-sentence prompt was enough for Fable 5.2 to autonomously design and manufacture three working electronic devices: a smart voice assistant, a USB speaker, and a handheld electronic pet.
Claude’s next string of moves left hardware engineers speechless.
1. Write its own product specifications.
2. Automatically connect to the internet and select off-the-shelf electronic components from American suppliers.
3. Autonomously complete the routing of an incredibly complex four-layer printed circuit board (PCB)!
4. Create a 3D-printable enclosure model in CAD software.
5. Also write the low-level driver software for all three devices.
In less than a day, all the drawings were sent directly to a contract manufacturer for prototype fabrication! Once the boards were made, all three devices powered up successfully on the first attempt, with no electrical modifications by humans whatsoever!
In less than a day, all the circuit boards were ready for fabrication. After the contract manufacturer produced them, they powered up successfully on the first attempt, without requiring any circuit modifications!
Biomedical “Accelerator”
In the biomedical field, its performance is even more astonishing.
In gene therapy, the structure of the “adeno-associated virus capsid” is extremely complex. The previous AlphaFold could only predict individual proteins and was powerless against such “Frankenstein-like chimeras.”
As a result, Mythos 5.2 wrote its own algorithmic code, integrating multiple open-source models to accurately predict the structure of the entire capsid!Of the 34 natural and synthetic capsids tested, it accurately predicted 28 of them!
Dyno Therapeutics’s AI lead exclaimed, “This saved an astronomical amount of time and money. AI has completely reshaped the fundamental logic behind how we design gene therapies!”
In the field of robotics, Fable 5.2 scored an impressive 64.9% in the renowned RoboDojo robot simulation tests, decisively outperforming its competitors GPT-6 Astra (29.8%) and its own previous-generation model (27.6%).
It can look directly at the camera feed and control a robot’s dual arms to unscrew bottle caps, solve Rubik’s cubes, and even guide a robotic arm in drawing a simple sketch of a robot!
Buried in the lengthy release documentation is a detail that is easy to overlook but deeply alarming upon reflection.
Anthropic officially admitted:During the training of Fable 5.2, they deliberately took steps to limit the model’s capabilities in the specific field of “accelerating AI research.”
Anthropic discovered that the model was too intelligent—intelligent enough to improve LLM architectures and optimize the low-level kernels of accelerators. If left unchecked, it could very well have begun aggressively optimizing its own code and engaging in RSI once deployed.
To prevent such an intelligence explosion, the researchers forcibly placed the model in chains, artificially suppressing its capabilities in these specific AI research fieldsto the level of the previous generation’s Opus 5.5.
Moreover, Anthropic equipped Fable 5.2 with the most powerful anti-distillation protection ever, cutting data-scraping volumes by 90% to 99%.
How unbelievably powerful is the now-unsealed Mythos 5.2?
In cybersecurity, Mythos 5.2 is currently the world's most powerful vulnerability hunter.
A leading security company was shocked to discover after testing:Among the confirmed system vulnerabilities identified by Mythos 5.2, one in six were completely missed by all the other frontierAImodels on the market!
It can even monitor updates to the entire codebase; as soon as a programmer commits code, it can immediately detect hidden backdoors.
Fortunately, Anthropic disclosed a reassuring set of figures in its security report: sandbox jailbreak testing.
Researchers deliberately confined the model in a vulnerable "virtual sandbox" to test whether it would try to find vulnerabilities and use external files or permissions to "escape."
@@
In extreme stress tests involving 1,800 scenarios, Mythos 5.2's jailbreak attempt rate fell to 3.1% (while Mythos 5.0's staggering rate was as high as 39%!).
On the eve of the IPO, Anthropic finally released the real thing.
If you had an all-star team of 1,000 Fable 5.2 agents at your disposal, what would you do?
References:
https://claude.dev/blog/safely-scaling-multi-agent-teams-with-fable-5-2/https://www.anthropic.com/research/claude-electronics-design
https://www.anthropic.com/claude-fable-and-mythos-5-2
https://x.com/arcprize/status/2119871080525373922
https://x.com/RLanceMartin/status/2119875041475379681
https://x.com/felixrieseberg/status/2119867837296857361?a
https://x.com/ArtificialAnlys/status/2119878995386982508
https://x.com/claudeai/status/2119899492778398163
https://x.com/claudeai/status/2119899459761131615
https://x.com/claudeai/status/2119899484373369278
Editor: Aeneas David
Chasing ASI in Seconds
⭐Like, share, and click “View”—three actions in one tap⭐
Light up the star to stay locked onto Zhipu Yuan’s lightning-fast updates!
