news.volyx.in

Claude Fable 5.1 and Claude Mythos 5.1 (anthropic.com)

1419 points by denysvitali · 41 days ago · 1395 comments on HN

Article summary

Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, the latest versions of their AI models for coding and knowledge work. Fable 5.1 is generally available and offers improved performance, reduced costs, and enhanced safeguards. The model has demonstrated its capabilities in various benchmarks and has received positive feedback from early-access partners. The new version also includes features such as Enterprise Frontier Safeguards, which provides customers with complete privacy and state-of-the-art prevention of adversarial use.

Main themes

  • AI model capabilities
  • Cost and pricing
  • Safeguards and restrictions
  • Technical domain and use case
  • User experience and feedback
  • Quota limits and transparency
  • Model improvement and development

What commenters say

  • Some users have had positive experiences with Fable, finding it to be very smart and capable of handling complex tasks, but others have found it to be too expensive and prone to downgrading to Opus.
  • Fable's safeguards can be overly restrictive, causing it to refuse to work on certain tasks, especially those related to security or cybersecurity.
  • The model's usefulness varies greatly depending on the technical domain and specific use case, with some users finding it to be very helpful and others experiencing frequent issues.
  • The cost and quota limits of Fable have been a point of contention, with some users feeling that the limits are too restrictive and others appreciating the reduced costs offered by the new version.
  • Some users have reported that Fable has improved over time, with the guards being less strict than they were initially, but others have continued to experience issues.
  • The lack of transparency and clarity around the quota limits and pricing has led to frustration and confusion among some users.
  • The model's performance and capabilities have been impressive in certain areas, such as planning and problem-solving, but have fallen short in others, such as reverse engineering and security-related tasks.