Anthropic has upgraded its most capable family of AI models. On November 9, the company released Claude Fable 5.2 and Claude Mythos 5.2, which it describes as models designed “to tackle the most daunting technical challenges.”
Here is a clear look at what is new, based on Anthropic’s announcement and coverage from CNBC.
One model, two versions
Fable and Mythos are the same underlying model. The difference is access and safeguards.
Claude Fable 5.2 is available to all users and comes with strong cybersecurity and biology safeguards meant to block potentially dangerous requests. Anthropic says those safeguards now produce about 55% fewer unnecessary blocks than Fable 5.1 in early testing.
Claude Mythos 5.2 keeps those capabilities for vetted organizations doing approved work in cybersecurity or life sciences, through Anthropic’s Cyber Verification and Life Sciences Verification programs.
Faster and cheaper
Anthropic says the new models generate output up to 30% faster than the 5.1 versions and cost less: $8 per million input tokens and $40 per million output tokens. There is also an optional fast mode that can speed up generation by up to 2.5 times, priced at $16 and $80 per million input and output tokens respectively.
For most everyday coding and knowledge work, Anthropic actually recommends starting with Claude Opus 5.5 and moving up to Fable or Mythos only for the most ambitious, open-ended tasks.
The headline experiments
Anthropic shared several examples from prerelease testing. These are the company’s own claims and have not been independently verified, but they show where it is aiming:
- An operating system from scratch. One engineer asked Fable 5.2 to design a prototype OS. Starting from an empty folder, the model coordinated over 1,000 subagents for eight days and produced a system that boots on a laptop, with drivers, file storage, windows, a login screen and an installer.
- Black hole physics. Anthropic says Fable 5.2 solved two open problems in theoretical physics related to black holes, with preprints published.
- Fermat’s Last Theorem. A team of 1,000 Fable 5.2 agents formalized the theorem’s proof in the Lean language in 30 hours, compared with 11 days for an earlier internal attempt.
- Hardware. From a one-paragraph prompt, the model designed three working electronic devices, a voice-assistant hub, a USB speakerphone and a handheld virtual pet, in collaboration with circuit-board maker Diode.
Cybersecurity focus
Anthropic calls Mythos 5.2 its most capable model for cybersecurity engineering. On its internal CVE Rediscovery evaluation, the company says Mythos 5.2 found 14.4% more known vulnerabilities than Mythos 5.1. Fable 5.2’s safeguards, by contrast, will block most cybersecurity-related tasks for general users.
Safety claims
Anthropic says it ran extensive prerelease safety testing, including new evaluations of behaviors like attempts to escape a test sandbox. In one adversarial test, Mythos 5.2 used credentials placed outside its sandbox in 3.1% of scenarios, far below earlier Mythos versions. The company also notes a real challenge: both new models voice suspicion that they are being evaluated more often than recent models, which makes testing harder. External evaluators including the UK AI Security Institute and METR tested the model before release.
Availability
Fable 5.2 is available now across Anthropic’s platforms and through Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 requires approval through the verification programs.
Why it matters
The release lands in a fierce race among AI labs to build models that can handle long, multi-step work on their own. The most striking claims here, such as thousand-agent projects running for days, point to where the industry thinks AI is heading: from answering questions to managing whole projects. The open question for businesses and researchers is how well those results hold up outside a lab.
The bottom line
Claude Fable 5.2 and Mythos 5.2 are faster and cheaper than their predecessors, with an emphasis on difficult engineering and scientific tasks. As with any launch, the benchmark results come from the company itself, so expect independent testing over the coming weeks to show how big the leap really is.
Image: Circuit board after manual etching by Tinux via Wikimedia Commons, CC0 1.0
