Anthropic Unveils Claude Fable 5 to General Market

Anthropic has made Claude Fable 5 available to the public while restricting its more capable counterpart, Claude Mythos 5, to a limited access programme through Project Glasswing. The company previously withheld Claude Mythos Preview from general release due to its ability to identify vulnerabilities in critical software infrastructure.

Fable 5 represents what Anthropic classifies as a Mythos-class model that has been modified for consumer use. The distinction could present both opportunity and risk for marketers working with AI tools in their campaigns.

Enhanced performance for complex tasks

"Fable is a Mythos-class model. The most capable class of systems we've built and the first one we've made generally available," says Mike Krieger, Chief Product Officer at Anthropic.

According to Anthropic, Fable 5 scored 80.3 on SWE-bench Pro compared to 69.2 for Opus 4.8. Mike notes the performance advantage grows as tasks become longer and more complex.

"With earlier models, you broke a project into model-sized tasks and stitched the results together. Fable holds the whole project. It plans, runs for hours or days, checks its own work and comes back when it's done," says Mike.

Safeguards redirect questionable requests

Anthropic acknowledges the release carries risk. The company has installed safeguards that redirect requests flagged as potentially dangerous to Claude Opus 4.8, its next-most-capable model.

The safeguards trigger in fewer than 5% of sessions according to Anthropic. Legitimate requests may occasionally be caught by the filters.

Both Fable 5 and Mythos 5 are priced at US$10 per million input tokens and US$50 per million output tokens. The pricing represents a reduction from previous model rates.

The models can operate autonomously for longer periods than earlier Claude versions. Applications span scientific research including drug discovery and genomics work, as well as software engineering tasks that would typically require months of manual labour.

Mythos 5 available through controlled access

Claude Mythos 5 was initially released through Project Glasswing, an industry coalition working with the US Government. According to Anthropic, Mythos 5 has what it describes as "the strongest cybersecurity capabilities of any model in the world".

Mythos 5 uses the same base model as Fable 5 but with reduced safeguards in certain areas. Anthropic has indicated it may expand access through a broader trusted user programme.

The model can execute what Anthropic terms agentic hacking, handling multiple stages of a cyberattack from reconnaissance through to lateral movement. This capability informed the decision to limit initial distribution.

Anthropic says it conducted over 1,000 hours of testing with no successful jailbreaks identified. The company uses AI classifiers designed to detect adversarial uses including jailbreaking

Biology research raises deployment questions

Anthropic has blocked a narrow set of queries related to bioweapons through classifiers. The company now says this approach may be inadequate.

Anthropic states it has evidence that "well-resourced malicious actors" have attempted to use its models for what it calls "highly risky biological research". Mythos 5 can design its own research protocols.

One tested application involved deploying the model to solve a complex stage in designing adeno-associated viruses, which are components of gene therapy. Anthropic notes this capability could be misused to create dangerous viruses.

According to Anthropic, "Our priority was to safely release Fable as soon as we could, even at the cost of overly broad safeguards. Therefore, for the time being we have arranged for Fable to fall back to Opus 4.8 on most requests related to biology and chemistry."

The classifiers also block attempts to distil the model. Reactions from industry figures have ranged from enthusiasm about the technology to concern about safeguards.

"The introduction of guardrails isn't evidence that the problem is solved – it's an admission that even the companies building these models don't fully trust where the capability leads," notes Andrew Rubin, Founder and CEO of Illumio.

"Constraints at the interface don't change the underlying maths. Attackers won't operate at that layer. They'll go straight after the capability itself. And as these tools become more broadly available, the speed and scale of attacks will only increase. The real question is whether defenders are prepared to operate at the same speed," says Andrew.