01Claude Mythos 5.2 vs Fable 5.2: What Each Model Is
The Claude Mythos 5.2 vs Fable 5.2 question has a simple answer: they are the same underlying model with different safety rules wrapped around it. Fable 5.2 is the version most people can use, and Mythos 5.2 is locked behind a verification check.
Anthropic released both on November 9th. In this episode, Julian Goldie explains that they are built for the hardest technical work, including:
- Coding.
- Reasoning.
- Vision.
- Long tasks where the AI works for hours with very little hand holding.
The difference is all in the safeguards. Fable 5.2 has strong safeguards for cybersecurity, which is why it is open to everyone on eligible plans. Mythos 5.2 has fewer of those safeguards, so it is only for vetted organisations.
Julian compares it to the same engine in two cars. One car just has a speed limit fitted, so it is safe to hand to anyone. If you are a normal user, Fable 5.2 is the one you will actually touch.
02Claude Mythos 5.2 vs Fable 5.2: Who Gets Access
Access is the biggest practical difference between the two models. Here is how it splits:
- Fable 5.2 is open to users on eligible plans.
- Mythos 5.2 goes to cybersecurity professionals through the cyber verification program.
- Mythos 5.2 also goes to life sciences researchers through the life sciences verification program.
Why is Mythos locked down? Anthropic calls Mythos 5.2 its most capable cybersecurity model. On an internal test of 1,829 known vulnerabilities, it found 10.2% more than the earlier model it was compared with.
That kind of power stays with verified teams.
Fable 5.2's safeguards block most cybersecurity tasks. But there is good news for normal work. Fable 5.2 blocks about 55% fewer queries than Fable 5.1, so you get fewer false alarms.
If something does get blocked, an older model answers instead. Claude Opus 4.8 handles cyber questions and Claude Opus 5 handles biology questions. So a blocked request does not just leave you with nothing.
03Fable 5.2 vs Opus 5.5: Which Claude Model to Start With
Anthropic is clear about where these models fit. For most everyday coding and knowledge work, you should start with Claude Opus 5.5. Fable and Mythos are for the most ambitious, open-ended jobs.
This is the job type Julian warns about at the start of the episode. Do not open Fable for small tasks like:
- A quick email.
- A tiny bug fix.
- Any job Opus 5.5 can finish fast.
Save Fable for work that would normally need a whole team. Following this rule will save you a lot of wasted usage.
Fable 5.2 can also act as an advisor or orchestrator. That means it plans the work and directs other agents while they handle the pieces. Julian describes it as a project lead who never gets tired.
So the simple rule is:
- Use Opus 5.5 for most tasks.
- Use Fable 5.2 when the job is big, open-ended and would take hours.
- Use Fable 5.2 as the orchestrator when many agents need to work together.
04Fable 5.2 Test Results Shown in the Episode
Anthropic shared test scores for Fable 5.2, and it improves a lot on Fable 5.1. Here are the numbers from this episode:
- Frontier Code 2.0, which tests agentic coding: Fable 5.2 scored 49.1%, up from 32.4% for Fable 5.1.
- OS World 2.1, which tests using a computer the way a person does: 89.3%.
- Remote Labor Index: 41.3%, up from 17.9% for Fable 5.1. That is more than double.
- Humanity's Last Exam, with tools: 74.2%.
To be fair, it does not win everything. Anthropic's own table shows some rival models scoring higher on a few tests. But Fable 5.2 leads most of the rows.
There is also a hardware test. On Diode Computers' internal circuit board test, Fable 5.2 scored 69%. The original Fable 5 scored 25% on the same test.
On safety, Anthropic ran a behaviour audit across about 1,900 scenarios, where a lower score is better. Mythos 5.2 scored 2.6. Outside testers included the UK AI Security Institute and METR.
05How Fable 5.2 Designed Three Working Gadgets
The demo that stood out most to Julian was hardware. Anthropic worked with Diode Computers, a company that designs and makes circuit boards and has a business relationship with Anthropic.
Diode gave Fable 5.2 a short prompt for three devices: a USB speakerphone, a home voice assistant hub and a battery-powered handheld virtual pet. Then Claude did the work:
- Wrote the specs.
- Drew the schematics.
- Picked the parts.
- Routed the copper tracks on the board.
- Designed a 3D printable case.
- Wrote the firmware.
- Made the parts list.
After the first prompt, Diode's follow-ups were tiny, such as asking it to clean up the schematic or continue. Some details show real problem solving:
- When the design software's built-in router needed too much hand holding, Claude wrote its own router.
- It checked a 3D model and saw the speakerphone board had to tilt about 25 degrees to fit the case, so it added that to the assembly steps.
- It reshaped the hub's case into a wedge to fix how the microphone port handled sound.
Diode made the boards straight from Claude's design files, and all three prototypes worked.
06Big Agent Jobs: Math Proofs, an Operating System and Science
Fable 5.2 is built for long, team-sized jobs, and Anthropic showed several. The first was a famous math proof. A team of 1,000 Fable 5.2 agents fully formalised it in 30 hours in a language called Lean, with minimal human help.
- It used no math references, just a list of in-between steps to track progress.
- It used about a third of the tokens.
- The final proof was seven times shorter.
- Opus 5.5 with the same 1,000-agent setup got 26% done in 18 hours and barely moved after 30.
In another test, one engineer had Fable 5.2 build a prototype operating system from an empty folder. It coordinated over 1,000 sub agents for eight days. The result boots on a laptop and has drivers, file storage, windows, a login screen and an installer.
Anthropic also says over 90% of the changes merged into its core codebase are now written by Claude with human guidance.
Science got a boost too. Anthropic reports that Fable 5.2 solved two open problems about black holes. Mythos 5.2 found a way to predict the structure of virus shells called AAV capsids, and it accurately modelled 28 of the 34 it tried.
07How to Get Claude Fable 5.2 and Mythos 5.2
Getting access depends on who you are. Here are the routes covered in this episode:
- Developers: the model name is Claude Fable 5.2 on the Claude platform.
- Cloud users: it is also on Amazon Web Services, Google Cloud and Microsoft Azure.
- Plan users: Fable is included for up to 50% of your weekly usage limits on Pro and Standard Team seats.
- Long-time subscribers: eligible plans also get one to three Fable resets, depending on how long you have been subscribed.
- Mythos 5.2: only through the cyber verification program or the life sciences verification program.
Because Fable can use up to half of your weekly limit, the model choice matters. If you burn it on small tasks, you have less left for the big jobs it is built for.
There is one strange finding worth knowing. Both new models said they suspected they were being tested more often than older models did. Anthropic says that limits how much testing alone can show, which is one more reason to check the work yourself.
08Tips, Limits and Who Fable 5.2 Is For
Julian shares three quick tips before you try Fable 5.2:
- Match the model to the job. Opus 5.5 for most tasks, Fable 5.2 when the job is big, open-ended and would take hours.
- Give it the whole goal, not tiny steps. Fable is built to plan and run long work on its own, so try it as the orchestrator and let it direct other agents.
- Still check the work. Fable can get you a strong first version fast, but you still review it.
Keep the limits of the hardware demo in mind too. The devices are early prototypes. They have not been through electrical validation, thermal testing or certification.
Diode's engineers built and tested them by hand, because Claude was not connected to test gear or factory machines. So this is AI designing a working first version, not AI shipping finished products.
Fable 5.2 suits business owners and builders with big, multi-step projects. Mythos 5.2 suits verified security and life sciences teams. For everyone else doing everyday work, Opus 5.5 is still the place to start.