Anthropic announced on the 2nd that it has unveiled its next-generation AI models, "Claude Fable 5.1" and "Claude Mythos 5.1." However, access to Mythos remains restricted.
Anthropic introduced "Claude Fable 5.1," an improved version of the existing "Claude Fable 5." This cutting-edge model delivers industry-leading intelligence for advanced multi-agent tasks involving coding and knowledge-based operations, significantly enhancing technical performance compared to its predecessor while simultaneously securing cost efficiency and safety for enterprise customers.
Anthropic has reduced the price for reading cached data in Fable 5.1 by 75% from previous levels. Reading cached data refers to the model re-reading input information that it has already processed and stored. This applies to both API (application programming interface) usage billed per token and any additional usage. The base API prices, excluding cache read costs, remain unchanged at $10 per 1 million input tokens and $50 per 1 million output tokens, the same as for Fable 5.
Alongside improvements in performance and cost, the precision of "safety mechanisms" has also been significantly strengthened. Based on customer feedback since the launch of Fable 5, Anthropic introduced a new data control feature called "Enterprise Frontier Safety." This ensures data privacy at a level where customer data is not stored at all while thoroughly preventing adversarial use of AI.
Furthermore, the precision of safety classifiers has been improved to prevent AI models from excessively blocking legitimate research requests in fields such as biology and cybersecurity. The latest biological safety mechanisms have reduced intervention frequency for normal requests by 85% compared to the previous version. For users of "Claude Code," the frequency of cyber safety interventions per session is expected to decrease by an average of about 60% compared to before.
Meanwhile, Fable 5.1 is currently available to all general users. The newly unveiled "Mythos 5.1" is provided only on a limited basis through Anthropic's trust-based access program.
in TECH
