Skip to main content

Claude 5 Arrives: AI That Codes Like a Pro and Cracks Scientific Mysteries

Anthropic Launches Next-Gen Claude AI Models

Artificial intelligence research company Anthropic has unveiled its fifth-generation Claude AI system, introducing two specialized models that push the boundaries of what language models can achieve. The release marks a significant leap forward in both general-purpose and specialized AI capabilities.

Image

Claude Fable5: The Coding Powerhouse

The general-purpose Fable5 model has demonstrated extraordinary performance across multiple benchmarks. Its coding abilities are particularly impressive:

  • Software Development: In the SWE-Bench Pro test that evaluates solving real GitHub issues, Fable5 scored 80.3%, outperforming its predecessor Claude Opus4.8 (69.2%) and GPT5.5 (58.6%).
  • Real-World Impact: Payment processor Stripe reported that what previously took five months of developer work was completed in days using Fable5. The model migrated a 50-million-line Ruby codebase in one day – work that would have taken a human team over two months.
  • Financial Analysis: Trading firms like IMC found the model nearly passed all their trading analysis tests, while its visual interpretation skills allow it to extract data from complex scientific charts with remarkable accuracy.

Claude Mythos5: The Science Specialist

While Fable5 focuses on general applications, Mythos5 targets specialized scientific research with fewer restrictions. Currently available only to select partners and the U.S. government through Project Glasswing, it's showing groundbreaking results:

  • Drug Discovery: In blind tests, Mythos5 autonomously designed effective candidate drugs for 9 out of 14 protein targets, accelerating the process tenfold.
  • Scientific Hypotheses: Researchers preferred Mythos5's molecular biology theories in 80% of cases, including a novel E. coli protein mechanism that's since been verified.
  • Genomic Research: Working continuously for a week without human input, the model analyzed single-cell data from 138 species, creating a machine learning model that outperformed one published in Science magazine – despite being 100 times smaller.

Balancing Power With Responsibility

These capabilities come with significant considerations:

  • Cost: Pricing starts at $10 per million input tokens and $50 per million output tokens – nearly double previous models.
  • Security Measures: Fable5 includes a novel classifier that redirects potentially dangerous queries (about 5% of requests) to less powerful models. After 1,000 hours of testing, no successful jailbreaks were found.
  • Data Retention: Anthropic now keeps user data for 30 days to monitor for new attack patterns.

Availability

Fable5 is currently accessible via the Claude API and enterprise plans, with gradual rollout to Claude.ai subscribers. Users can test it free until June 22 before usage credits apply.

Key Points

  • Fable5 excels in coding, completing months of work in days
  • Mythos5 proposes valid scientific hypotheses and accelerates drug discovery
  • New models cost nearly twice as much as previous versions
  • Strict security measures prevent misuse of advanced capabilities
  • Available now through API, with wider rollout underway