Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, enhancing capabilities in programming and scientific research.
Anthropic has introduced two new AI models, Claude Fable 5.1 and Claude Mythos 5.1, touted as leading tools for programming and scientific research.
Claude Fable 5.1 is available for all users, while Claude Mythos 5.1 is limited to select partners. The latter features enhanced security measures tailored for cybersecurity and biological applications.
Pricing for Fable 5.1 has dropped by an average of 25%, with savings reaching up to 45% for complex tasks, thanks to reduced data reading costs from cache.
Additionally, the company plans to launch its Enterprise Frontier Safeguards (EFS) security system this fall, allowing corporate clients to keep their data in private cloud environments for improved confidentiality.
In performance tests, Fable 5.1 outshone its predecessors in coding and logic. Notably, it successfully diagnosed a rare system failure at Millennium, an investment firm, a challenge that had perplexed engineers for five years.
In scientific applications, the AI achieved significant milestones, including:
- Mapping: Creating a high-resolution map of Venus from 30-year-old NASA data, enhancing elevation accuracy by 25%;
- Biology: Designing protein structures that bind to targets ten times more effectively than current options;
- Optimization: Boosting the performance of seven biological models by 2.5 times within days through code optimization for graphics processing units.
Moreover, Anthropic has refined its security filters, reducing the blocking of harmless vulnerability-seeking requests by 60%. While the AI can identify security flaws, it is prohibited from assisting in creating them.
In August, the company raised concerns about trust, deception, and collusion in multi-agent AI systems.
