Anthropic Releases Claude Fable 5.2 and Mythos 5.2, Topping the...

Just now, Claude Fable 5.2 topped the world rankings, slaying Opus 5.5 overnight. ifanr Download the app for exclusive early access. November 10, 2026, 06:42:02, from Beijing. Updated November 10, 2026, 06:52:25

1,000 AI Agents completed a formal proof of Fermat's Last Theorem in 30 hours, and worked continuously for 8 days to build from scratch an operating system that can run on a laptop.

Just now, Anthropic officially released two new models, Claude Fable 5.2 and Mythos 5.2. In addition to the achievements above, the new models also solved two black hole physics problems, increased generation speed by up to 30%, and lowered API prices by 20%.

Interestingly, it has been only two months since Anthropic CEO Dario Amodei publicly called for slowing down AI development, and Anthropic has once again launched a stronger frontier model. At the same time, Anthropic is preparing for an IPO valued at as much as $1 trillion to $1.5 trillion.

Calling for slower AI on one hand while sprinting toward a trillion-dollar IPO on the other—what exactly has Anthropic brought out this time?

1,000 Agents worked for 8 days straight and built an operating system from scratch

According to Anthropic's positioning, Claude Fable 5.2 and Mythos 5.2 are flagship models aimed specifically at complex technical challenges, with key improvements in coding, reasoning, visual understanding, and scientific research capabilities.

The two use the same underlying model, and the main difference lies in safety restrictions. Fable 5.2 adds anti-abuse mechanisms for cybersecurity and biological research and is open to ordinary users; Mythos 5.2 retains more complete professional capabilities and is mainly provided to vetted institutions.

Judging from the officially published benchmarks, the new models perform quite impressively.

In the FrontierCode 2.0 test, which measures AI Agent coding ability, Fable 5.2 scored 49.1%, surpassing Claude Opus 5.5's 45.5%, GPT-6 Astra's 42.8%, and Gemini 4 Carbon's 41.1%.

In the FrontierSWE v2 software engineering test, Fable 5.2 scored 69.9%, an increase of 13.6 percentage points over the previous generation. And in the FrontierCode 2.0 test, Fable 5.2 achieved results close to Opus 5.5's highest reasoning intensity with lower reasoning intensity, at a per-task cost of about half that of the latter.

The latest rankings from third-party evaluator Artificial Analysis also confirm this.

Claude Fable 5.2 topped Intelligence Index v5 with 65 points, leading Gemini 4 Carbon by 5 points and Claude Opus 5.5 by 6 points, and took first place in 5 of 9 tests. Its main advantages are concentrated in scientific reasoning, knowledge understanding, and coding ability. However, the average per-task cost in the strongest reasoning mode reached $26.5, about 2.6 times that of Opus 5.5.

However, more noteworthy than the benchmark scores are the actual engineering cases Anthropic demonstrated.

An engineer asked Fable 5.2 to design an operating system from scratch. Starting from an empty folder, the model autonomously coordinated more than 1,000 sub-Agents, worked continuously for 8 days, and completed code writing, testing, and review.

In the end, Claude successfully developed an operating system that can boot and run on a laptop.

According to the official demonstration, the system has its own kernel, supports disk, keyboard, and network drivers, and also features file storage, a windowed interface, a login screen, an installer, and built-in Claude functionality.

Although it is still an experimental prototype system, the fact that AI can autonomously coordinate thousands of Agents to complete such a complex software engineering project is enough to show how its capabilities have changed.

In another practical case, an early version of Fable 5.2 also migrated a key Anthropic production service from Python to Rust, increasing its startup speed to about 26 times the original.

After the release, AI entrepreneur Matt Shumer commented that Fable 5.2 is the strongest model he has ever used and can even complete complex projects that other models found difficult in the past, but it sometimes is "over-responsible" and even obstructs normal work tasks.

Netizen Kevin Ngo also showed off a 3D game made with Fable 5.2 that can be played directly in the browser, in which players can ride a boat through 8 dreamlike worlds;

Another developer, Flavio Adamo, also showcased 3D works generated by the model, exclaiming that its 3D creation ability is astonishingly strong.

In addition, netizen Synopsis also used the classic "pelican riding a bicycle" test to challenge Fable 5.2, posted the work generated by the model, and directly called it the new-generation BOSS.

Anthropic also revealed that currently, in the company's core codebase, more than 90% of merged code changes are written by Claude under human guidance.

From fixing simple bugs to autonomously planning and executing large engineering tasks lasting several days, AI programming is gradually gaining the ability to independently undertake complex projects.

Completing a formal proof of Fermat's Last Theorem in 30 hours, and also solving difficult problems in black hole physics

If the improvement in programming ability is still within expectations, then the scientific research results announced by Anthropic are even more surprising.

Fermat's Last Theorem is one of the most famous problems in the history of mathematics. Although mathematician Andrew Wiles completed the proof as early as the 1990s, converting it into a formal proof that a computer can verify step by step is still an extremely complex task.

Anthropic previously used an unreleased Claude research model and, under the guidance of human researchers, spent 11 days completing the related work.

Now, a team composed of 1,000 Fable 5.2 Agents completed the formal proof of Fermat's Last Theorem in only 30 hours, with almost no manual intervention.

The model established collaboration mechanisms on its own during the experiment, and the generated proof was shortened to one-seventh of its previous length, while the number of operation Tokens consumed was also reduced to one-third of the original. As a control, Claude Opus 5.5 under the same configuration ran for 30 hours and completed only about 27% of the proof work.

In addition to mathematics, Fable 5.2 also solved two open problems in theoretical physics related to black holes.

One of them involves the phase transition mechanism when black holes form. Scientists discovered as early as 1983 that black holes in a specific theoretical model would suddenly form as temperature changes, but whether this conclusion can be extended to smaller quantum scales has long lacked complete calculation results.

Claude basically autonomously completed calculations that had long remained unfinished and obtained the same abrupt-transition conclusion. In another study, the model determined the precise applicable boundary of a certain analytical method in quantum gravity theory.

At the same time, Mythos 5.2 also combined two existing open-source protein structure prediction models to develop a new method for predicting the structure of complete viral capsids, and among the 34 capsids tested, it successfully and accurately modeled 28 of them.

Designing three electronic devices in one go, and robots are also beginning to understand instructions

Claude's capabilities are also beginning to extend to hardware design in the real world.

In an experiment in which Anthropic collaborated with the circuit board design and manufacturing company Diode, researchers used only a single prompt to have Fable 5.2 design three electronic devices that can actually operate, namely a home voice assistant hub, a USB conference speaker, and a handheld electronic pet.

Throughout the process, Claude autonomously wrote product specifications, selected electronic components, designed four-layer PCB circuit boards, built 3D-printed enclosure models, and developed supporting software.

According to Anthropic, the circuit board design of each device reached a manufacturable state in less than a day. After Diode completed manufacturing, Claude then guided engineers in testing and assembly, and ultimately all three devices operated successfully without the need to modify the electrical design.

In terms of robot control, Fable 5.2's progress is equally evident.

In the RoboDojo Sim dual-arm robot simulation test, Fable 5.2 achieved a success rate of 64.9%, surpassing the 27.6% of the previous generation Fable 5.1 and the 29.8% of GPT-6 Astra.

In the official demonstration, a physical robot controlled by Claude also completed tasks such as unscrewing a bottle cap, solving a Rubik's Cube, and drawing a robot pattern.

Coincidentally, netizen Tekon Burns also used a real Rubik's Cube to test Fable 5.2. After uploading only 6 photos, the model provided a solution with very few steps, and it did not follow the common human solving approach of first forming the white cross and then restoring the white face.

From text and code on the screen to circuit boards and robots in reality, Claude is trying to transform its reasoning ability into actual physical operations.

Compared with Fable 5.2, the focus of Claude Mythos 5.2 remains cybersecurity.

Anthropic first launched Mythos in April this year. Considering its powerful vulnerability discovery and exploitation capabilities, it was initially opened only to selected institutions, and only later was the Fable series launched for public use with added security restrictions.

In this upgrade, Mythos 5.2 was able to rediscover 89.6% of known software vulnerabilities in the internal CVE Rediscovery test, higher than the 78.3% of the previous generation.

A security company also found that among the vulnerabilities confirmed by Mythos 5.2, about one-sixth were not discovered by other frontier models it tested.

As capabilities improved, Anthropic also further strengthened its model safety measures. In a sandbox escape test containing more than 1,800 scenarios, the proportion of Mythos 5.2 using credentials or files outside the sandbox dropped to 3.1%, a sharp decline from 21% in the previous generation.

But more noteworthy is the proactive restriction Anthropic placed on the new model.

Officials explicitly stated that during training, the new model's capability improvements were specifically restricted in areas such as AI accelerator kernel optimization and large language model architecture improvement.

Therefore, although the overall capabilities of Fable 5.2 and Mythos 5.2 are significantly enhanced, their performance on these tasks that could directly accelerate next-generation AI R&D remains roughly at the level of Claude Opus 5.5.

Anthropic emphasized that the relevant adjustments did not affect capabilities such as ordinary Agent programming, hardware development, and traditional machine learning research.

In addition, Anthropic strengthened measures to prevent illegal distillation of model capabilities, claiming that the relevant technology can reduce the amount of data obtained by malicious distillers by 90% to 99%.

Price cut by 20%, and Claude Pro also has hidden benefits

Claude Fable 5.2 officially launched today, covering the Claude platform as well as cloud services such as AWS, Google Cloud, and Microsoft Azure. Developers can call the new model through claude-fable-5-2.

Compared with the previous generation, the standard API price of the new model is reduced by about 20%, charging $8 per million input tokens and $40 per million output tokens.

The two models also introduced an optional Fast Mode, with token generation speed up to 2.5 times that of standard mode, while the price rises accordingly to $16 per million input tokens and $80 per million output tokens.

Anthropic stated that the Fable series is mainly aimed at complex engineering, mathematics, and scientific research tasks, while for everyday programming and knowledge work it still recommends prioritizing Claude Opus 5.5.

To promote the new models, Anthropic also launched a limited-time trial campaign.

From November 9 to 16 local time, eligible Claude Pro users and Team standard seat users can use the Fable series models within 50% of their original weekly usage quota without purchasing additional usage quota, and receive one free quota reset opportunity.

Max, Ultra, and some Team and Enterprise premium seat users can receive 1 to 3 usage quota resets depending on the length of their paid subscription.

It should be noted that the reset function only restores quota that has already been consumed and does not permanently increase the usage limit of the subscription plan. As for Mythos 5.2, it is currently still mainly open to verified cybersecurity institutions and life science research organizations.

Calling for slowing down AI on one hand, sprinting toward a trillion-dollar IPO on the other

The timing of Anthropic's release of the new models this time is quite delicate.

In September this year, Anthropic CEO Dario Amodei publicly called for slowing the pace of AI development, hoping the entire industry would handle rapidly improving model capabilities and their potential risks more cautiously.

Now, just two months later, Anthropic has launched a new generation of frontier models with stronger capabilities.

In response, Dianne Penn, head of research product management at Anthropic, said in an interview with CNBC, "What we called for is controlling the pace of development, not pausing development."

Anthropic stated that the new models underwent extensive safety and alignment testing before official release, with external institutions such as the UK AI Safety Institute and METR participating in the evaluation.

However, investment in safety is also facing increasingly realistic commercial pressure.

Now, Anthropic is making final preparations for an IPO, and management is about to launch the listing roadshow, with a target valuation of $1 trillion to $1.5 trillion, lower than the previously rumored market target of $2 trillion.

According to Anthropic's updated prospectus, as of September 30 this year, Anthropic achieved revenue of $33.24 billion, a year-on-year increase of about 1,200%. For a company about to sprint toward a trillion-dollar IPO, technological leadership means greater business opportunities, and also an increasingly unbearable cost of slowing down.

"Special statement: The above content (including videos, pictures, and audio if any) is uploaded and published by users of "Dafeng Hao," a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."