A key technical milestone of this release is Fable 5.1’s ability to discover software vulnerabilities and solve long-term engineering problems, such as identifying the root cause of rare system crashes that have historically eluded human engineers.
The update introduces significant changes to AI infrastructure costs and data privacy.
Anthropic claims Fable 5.1 reduces expenses for typical workloads by 25% compared to its predecessor, with savings reaching 45% for "agentic" tasks—work where the AI autonomously uses tools and processes heavy amounts of data.
This is achieved by cutting the price of cache reads, which occur when the model reuses previously processed information.
Furthermore, a new "Enterprise Frontier Safeguards" system allows corporate clients to maintain total data privacy by storing information on their own cloud infrastructure rather than Anthropic's servers, effectively providing a zero data retention policy.
Looking ahead, the Mythos 5.1 variant will serve as a specialized tool for high-stakes scientific discovery under government-partnered oversight.
In early testing, the model designed protein binders with 10 times the affinity of top human-led designs and optimized genomic research models to run up to 2.5 times faster on hardware like NVIDIA H100 GPUs.
While these capabilities are currently limited to vetted US organizations, Anthropic plans to expand access to a broader group of international scientists and researchers.
To comply with the EU AI Act, the company is also implementing an invisible watermark on text outputs and launching a detection tool for regulators and law enforcement.