The strongest AI model Anthropic has ever shipped is now open to all. It's also built to say no.
The model is called Claude Fable 5, and it went live this week. In some high-risk areas, it just refuses to answer.
Fable 5 is the first public version of Mythos, the firm's top model. It's strong at writing code, doing office work, and reading images.
But it won't touch a few topics. These include cybersecurity, biology, and chemistry.
In those areas, it blocks the answer. It hands the question to an older, safer model called Opus 4.8.
The firm says that handoff is rare. Early data shows at least 95% of chats run start to finish on Fable.
We cover the business side of AI moves like this every morning in Market Briefs, five minutes a day, plus a free investing masterclass when you sign up.
Mythos first showed up as a preview in April. At first, only a few partners could use it.
Last week the firm widened access to hundreds of groups in 15 countries. Most of them run key systems, like power and water.
Now a public version is open to anyone. The firm also tried hard to break its own safety net first.
A paid bug hunt ran more than 1,000 hours and found no broad way past the limits. Outside testers tried too, and also came up short.
Power costs money here. Fable 5 runs $10 per million tokens of input and $50 per million of output.
Tokens are just the small chunks of text an AI reads and writes. That price is double what Opus 4.8 costs.
There's also a catch on your data. The firm now keeps 30 days of all traffic.
That holds even for clients who used to have no-keep deals. The firm says it won't use the data to train the model.
It says the data only helps it catch new attacks. It also helps cut down on false alarms.
This isn't happening in a quiet corner. The firm is getting ready to go public, next to OpenAI and Elon Musk's SpaceX.
It also comes right after a warning from Anthropic itself. The firm asked AI labs to agree on a shared "brake pedal" for the most advanced systems.
It even warned that AI may soon start to improve itself, with no human in the loop. So the same company warning about speed just shipped its fastest model yet.
Outside testers gave Fable strong marks. One data firm, Hex, said it was the first model to score 90% on its hardest test.
A coding tool called Base44 said Fable is great at building whole apps in one shot. Another, Genspark, said it beat every rival on design and game work.
The open question is cost. Many firms are already shocked by their AI bills, and Fable is the priciest choice on the menu.
Whether the extra power is worth double the price is the bet every customer now has to make.
Want to follow the AI money without the jargon? Join 350,000+ investors at Market Briefs and get a free 45-minute investing course when you join.