Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Faster, Cheaper AI for the Hardest Technical Jobs

Anthropic has released its most powerful artificial intelligence models yet. On Monday, 9 November 2026, the company launched Claude Fable 5.2 and Claude Mythos 5.2 — two versions of the same new model, built for the hardest technical work, from writing complex software to finding security flaws and designing electronic devices. The new models are faster than the ones they replace, and they cost 20 percent less to use.

The launch is the latest step in a fast-moving race between the world’s biggest AI companies. Anthropic, the company behind the Claude family of AI assistants, is pitching Fable 5.2 as its best model for “daunting technical work” — the kind of long, complicated jobs that earlier AI systems could not finish on their own. At the same time, it is keeping the most sensitive version, Mythos 5.2, behind a locked door, available only to vetted security and science organisations.

What Anthropic announced

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026. Both are updates to the Fable 5 and Mythos 5 models the company released in June 2026, with a smaller 5.1 update following in September. The new versions also ended weeks of online speculation about a version called “5.5”, which never appeared — Anthropic announced 5.2 instead.

The company describes the two new models as its most capable yet for technical work, setting “a new bar in coding, finance, and the sciences”. In plain terms, they are designed for jobs that need long, careful thinking: writing and fixing large amounts of software code, working through difficult maths and science problems, and carrying out projects that run for hours or even days.

Crucially, Fable 5.2 and Mythos 5.2 are the same underlying model. The difference is not how smart they are, but what they are allowed to do — and who is allowed to use them. Fable is the general version, open to all customers. Mythos is the restricted version, with its safety limits loosened so that approved experts can use it for cybersecurity and biology research.

One model, two versions: Fable vs Mythos, explained

To understand the launch, it helps to understand why Anthropic sells one model under two names.

Fable 5.2 is the version most people and businesses will use. It comes with strong safeguards for cybersecurity and biology. Those safeguards block requests that could be used to cause harm — for example, penetration testing, making software exploits, or most biology research. Because those protections are built in, Anthropic can offer Fable to everyone.

Mythos 5.2 is the same model with those safeguards relaxed. It is meant for people doing legitimate defensive work: security teams hunting for weaknesses in software before criminals find them, and life scientists doing approved biomedical research. Because the same abilities could be misused, direct access to Mythos is limited to organisations that pass Anthropic’s checks through its Cyber Verification and Life Sciences Verification programmes.

There is one exception. Business customers on Claude Enterprise can use Mythos’s vulnerability-finding abilities through a product called Claude Security, which scans their own code for weaknesses — without joining the verification programmes.

This split is not new. Anthropic has used the same approach since it launched Fable 5 and Mythos 5 together in June 2026. What is new is the scale of the improvement, and the price cut that comes with it.

What got better: the test results

AI companies test their models on standard tasks called benchmarks, so that performance can be compared. Anthropic’s own test results — reported by the company itself, not independently checked — show clear gains over the previous versions.

On coding, Fable 5.2 scored 49.1 percent on a test called FrontierCode 2.0 at its highest effort setting, compared with 32.4 percent for Fable 5.1. That test checks whether the people who maintain open-source software projects would accept AI-written changes to their code — in other words, whether the AI’s work is good enough for real use. Anthropic also said that Fable 5.2 at its “High effort” setting matched the performance of its daily-driver model, Opus 5.5, at its highest setting, at roughly half the cost per task.

On security, Mythos 5.2 found 89.6 percent of the weaknesses in an internal test called CVE Rediscovery, up from 78.3 percent for Mythos 5.1. The test used 1,829 previously disclosed software vulnerabilities and checked whether the model could independently rediscover flaws in software as it existed before those flaws were fixed. Anthropic said Mythos 5.2 was also better at reproducing vulnerabilities and developing fixes for them.

The models also handle much larger jobs. Fable 5.2 supports a context window of one million tokens — a token is roughly a word or part of a word, so that is an enormous amount of text, code or data it can hold in mind at once — and it can produce up to 128,000 tokens of output in a single response.

Anthropic says the new models generate output up to 30 percent faster than Fable 5.1 and Mythos 5.1, and that testers found they communicate more clearly than the earlier versions.

Cheaper to use: the new prices

For businesses that pay to use AI through an API — a connection that lets their own software talk to Anthropic’s models — the launch brings a welcome price cut.

Fable 5.2 and Mythos 5.2 both cost 8 dollars per million input tokens and 40 dollars per million output tokens. That is down from 10 dollars and 50 dollars for the 5.1 versions — a 20 percent reduction on both rates.

There is also a new optional “fast mode” that speeds up token generation by up to 2.5 times. It costs 16 dollars per million input tokens and 80 dollars per million output tokens — double the standard price, for up to two and a half times the speed.

Anthropic’s rivals charge more on their published price lists. According to a comparison by the financial news site Securities.io, OpenAI’s GPT-6 Astra lists standard prices of 10 dollars per million input tokens and 50 dollars per million output tokens — making Anthropic’s new rates 20 percent lower on both counts. That comparison is about the price of tokens, not the total cost of finishing a real task, which also depends on how many tokens a model needs and how often it gets the answer right.

For subscribers, Anthropic is also running a launch-week promotion: people on eligible paid plans get extra Fable usage, with Fable included for up to 50 percent of weekly usage limits for Pro plans and standard seats on the Team plan.

What the new models can actually do

Benchmarks are useful, but the most striking claims from Anthropic are about real projects its engineers tried with the new models.

In one internal experiment, an Anthropic engineer asked Fable 5.2 to build a prototype operating system — the basic software that runs a computer. The model coordinated more than 1,000 smaller AI agents working together over eight days, and produced software that could actually boot on a laptop. It is a glimpse of what AI companies mean when they talk about “agents”: AI systems that do not just answer questions, but carry out long, multi-step jobs on their own.

In a separate test, Anthropic asked Fable 5.2 to design electronic devices from start to finish, working with a company called Diode Computers that designs and makes circuit boards. The model designed three devices — a speakerphone, a voice-assistant hub, and a handheld virtual pet — and all three prototypes worked without any fixes to their circuitry. On a test of circuit-board layout called PCB Gym, Fable 5.2 scored 69 percent, compared with 25 percent for the original Fable 5. The devices are still early prototypes and have not been through electrical validation or certification, but the result shows how far AI has moved beyond writing text.

These are Anthropic’s own experiments, and the company has an interest in showing its models in the best light. Even so, independent analysts quoted by Computerworld said the improvements could make long-running AI agents practical for real business work, such as moving old software systems to new platforms and carrying out scientific research.

What it means for businesses and developers

For companies, the launch matters for three reasons: capability, cost and competition.

On capability, the new models are aimed at the hardest, longest jobs — the ones where earlier AI systems would lose track, make mistakes or simply give up. If Fable 5.2 really can sustain work over days, as Anthropic claims, it could change how software teams handle big migrations, how research labs run experiments, and how security teams hunt for weaknesses.

On cost, the 20 percent price cut — combined with faster output — makes these powerful models cheaper to run. Lian Jye Su, chief analyst at the research firm Omdia, told Computerworld that higher task completion rates could further reduce deployment costs, because a model that finishes a job correctly the first time is cheaper than one that needs several attempts. But Sanchit Vir Gogia, chief analyst at Greyhound Research, offered a note of caution: “Anthropic has cut the price of a token, not the price of a thought.” His point is that businesses should measure the cost of completed tasks, not just the cost of tokens.

On competition, the launch is part of a wider battle. Anthropic’s own benchmark table compares Fable 5.2 against its own Opus 5.5 and against rivals, including OpenAI’s GPT-6 Astra and Google’s Gemini 4 Carbon. Each company is trying to win over the developers and businesses who will build the next generation of AI-powered products — and price, speed and reliability are the battlegrounds.

The safety questions

With every more powerful AI release come the same questions: what could go wrong, and who is watching?

The first question is about who can use the most sensitive abilities. Mythos 5.2’s skill at finding software vulnerabilities is a double-edged sword. In the hands of defenders, it helps fix weaknesses before criminals exploit them. In the hands of attackers, it could help them find weaknesses faster. Pareekh Jain, chief executive of Pareekh Consulting, told Computerworld that this “creates an AI-powered race between attackers and defenders”. Anthropic’s answer is its verification programmes: only vetted organisations get direct access to Mythos.

The second question is about data. Anthropic said customers who qualify for zero data retention can keep those arrangements, while other customers face 30-day retention of their data for safety monitoring. The company has also introduced Enterprise Frontier Safeguards, currently in private testing, which lets automated safety checks run on prompts and responses stored inside a customer’s own systems. Analysts warn that the retention rules could matter more than the price cut for banks, hospitals and other regulated businesses, and that procurement teams should get written confirmation of their arrangements before deploying the models for sensitive work.

The third question is about reliability. Even Anthropic’s own tests found quirks: the new models voiced suspicion that they were being tested more often than earlier models did. And analysts stress that reliability is still a problem without human supervision. “Reliability remains an issue without human supervision,” Omdia’s Lian Jye Su told Computerworld. His advice to businesses: monitor what AI agents do, run them in properly isolated environments, and keep critical changes subject to human approval.

What happens next

Fable 5.2 is available now on all of Anthropic’s platforms, and the launch-week promotion for paid subscribers runs this week. Security teams can apply for Mythos 5.2 through Anthropic’s verification programmes.

What to watch in the coming weeks:

Real-world results. Anthropic’s benchmarks are its own. Independent tests, and the experience of developers using Fable 5.2 on real projects, will show whether the gains hold up outside the lab.

The competition’s response. How OpenAI and Google answer a faster, cheaper Fable will shape prices and features across the whole industry.

The safety debate. As AI models get better at finding security flaws and designing hardware, governments and regulators are paying closer attention. Anthropic’s careful split between Fable and Mythos is part of that conversation — and it is unlikely to be the last word.

Frequently asked questions

What is Claude Fable 5.2?

It is Anthropic’s newest and most capable AI model for technical work, released on 9 November 2026. It is designed for demanding jobs such as complex coding, scientific research and engineering, and it is broadly available to all customers.

What is Claude Mythos 5.2, and how is it different?

Mythos 5.2 is the same underlying model as Fable 5.2, but with its cybersecurity and biology safeguards relaxed. It is the company’s most capable model for cybersecurity and life sciences research, and direct access is restricted to vetted organisations through Anthropic’s verification programmes.

How much does Fable 5.2 cost?

It costs 8 dollars per million input tokens and 40 dollars per million output tokens through Anthropic’s API — down from 10 and 50 dollars for Fable 5.1. An optional fast mode costs 16 and 80 dollars and speeds up token generation by up to 2.5 times.

Is Fable 5.2 better than Opus 5.5?

Anthropic says Fable 5.2 is a step above Opus 5.5 in almost all the capabilities it evaluated, and that Fable 5.2 at its “High effort” setting matched Opus 5.5 at its highest setting at roughly half the cost per task. Even so, Anthropic still recommends Opus 5.5 as the starting point for most everyday coding and knowledge work, with Fable reserved for the most demanding jobs.

Where can I use Fable 5.2?

It is available on the Claude Platform as “claude-fable-5-2”, and on Amazon Web Services, Google Cloud and Microsoft Azure. It is included for Pro, Max, Ultra, Team and Enterprise subscribers.

What was the rumoured “5.5” version?

Before the launch, online leaks suggested Anthropic would release versions called Fable 5.5 and Mythos 5.5. That did not happen: the company announced 5.2 instead.

Conclusion

Claude Fable 5.2 and Mythos 5.2 are a significant step for Anthropic: faster, cheaper and more capable than the models they replace, with test results that point to real gains in coding, security research and even hardware design. The price cut makes top-tier AI more affordable, and the launch-week promotion will put the new models in the hands of paying users quickly.

But the launch also underlines the trade-offs that now come with every powerful AI release. The most sensitive abilities are kept behind verification checks. Data retention rules will matter to regulated businesses. And reliability without human supervision remains an open question, even by the admission of the analysts who praise the improvements.

For now, the models are out, the prices are lower, and the race between the AI labs is faster than ever. The real verdict will come not from Anthropic’s own test scores, but from what developers and businesses manage to build with Fable 5.2 in the weeks ahead.

Sources

• Anthropic, “Introducing Claude Fable 5.2 and Claude Mythos 5.2”, 9 November 2026: https://www.anthropic.com/claude-fable-and-mythos-5-2
• Computerworld (InfoWorld), “Anthropic unveils Claude Fable 5.2, Mythos 5.2 with coding gains and lower prices”, 10 November 2026: https://www.computerworld.com/article/4246693/anthropic-unveils-claude-fable-5-2-mythos-5-2-with-coding-gains-and-lower-prices-2.html
• Securities.io, “Anthropic Launches Claude Fable and Mythos 5.2 at Lower API Prices”, 9 November 2026: https://www.securities.io/anthropic-launches-claude-fable-and-mythos-5-2-at-lower-api-prices/
• Anthropic, “Claude can design electronic prototypes from start to finish”, 9 November 2026: https://www.anthropic.com/research/claude-electronics-design

This article is for general information and education only. It is not legal or financial advice.