Anthropic has just announced the launch of Claude Fable 5, its most advanced model to date, and one it is tremendously proud of. This is how the company introduced it, in fact: “Fable 5’s capabilities surpass those of any other model we have brought to market. It leads in nearly every AI performance benchmark, demonstrating exceptional results in software engineering, knowledge management, computer vision, scientific research, and many other areas. The longer and more complex the task, the greater Fable 5’s advantage over our other models.
Overall, Fable’s introduction is, as expected, impressive. It includes the usual side-by-side performance metrics across different areas that Google, ChatGPT, and Anthropic itself have accustomed us to over the past few years, while highlighting its immense capabilities in fields as diverse and complex as software engineering, advanced vision, and even drug design.
To illustrate some of these capabilities, Anthropic points to examples such as:
So far, so good. Very good, in fact. Hats off to Fable.
But that introduction also includes a warning, in true Spider-Man fashion (remember the line, “with great power comes great responsibility?”):
“Launching such a powerful model comes with risks. Without safeguards, Fable 5’s capabilities in areas such as cybersecurity could be misused to cause serious harm. That is why we have launched the model with safety measures under which queries on certain topics will be answered by our next most advanced model, Claude Opus 4.8.
To launch the model safely and quickly, we tuned these safety measures conservatively: they will occasionally flag harmless requests, although they are triggered in fewer than 5% of sessions on average. With even more advanced models arriving in the coming months, we are working to improve our safety measures and reduce false positives as quickly as possible.”
You may remember that in April, Anthropic made headlines in mainstream media around the world following the launch (in a very, very limited form) of Mythos, its most advanced model, which had particularly sensitive capabilities in areas such as cybersecurity. At the time, Mythos’s arrival was seen as a double-edged sword: a tool capable of strengthening the cybersecurity of whoever used it, but also a threat if it fell into the hands of cybercriminals. Some governments even came to view it as a matter of national security, given its potential to affect the stability of strategic digital infrastructure.
As it turns out, Claude Fable 5 is “a Mythos-class model”. What does that mean?
Both Fable 5 and Mythos 5 are based on the same underlying model, but they differ in their safety systems and target audiences. Fable 5 is the commercial version open to the general public, businesses, and cloud platforms, and as we have seen, it includes safety measures that “censor” high-risk requests in real time and automatically redirect them to a lower-tier, safer model (such as Claude Opus) to prevent misuse in critical areas such as cybersecurity.
By contrast, Mythos 5 is a version without these safety restrictions in sensitive disciplines such as cybersecurity or life sciences, designed specifically for environments where commercial filters would hinder research. Because of the potential of these uncensored capabilities, access is restricted and controlled under the Project Glasswing program, making it exclusively available to government partners, biomedical researchers, and certified security experts operating under strict monitoring.
Both Fable 5 and Mythos 5 start at $10 per million input tokens and $50 per million output tokens, which is less than half the price of Claude Mythos Preview, the version unveiled in April that raised so many concerns.
Fable 5 is available starting today in Enterprise plans based on the Claude API. For subscription plans, through June 22, Fable 5 is included in Pro, Max, Team, and Enterprise plans at no additional cost. To use it after that date, usage credits will be required.
Image: ChatGPT and Claude
Your email address will not be published. Required fields are marked *
Δ