Anthropic has released Claude Fable 5, its most advanced public AI model to date, previously labeled as too dangerous for public access. The model belongs to a new "Mythos-class" tier, which Anthropic had initially kept internal due to safety concerns.
Claude Fable 5 shows significant improvements over prior models in benchmarks against GPT-5.5 and Opus 4.8, particularly in coding and vision tasks. Key performance metrics include strong scores on SWE-Bench Pro and FrontierCode, along with a notable demonstration of handling a 50-million-line code migration for Stripe.
The company has also implemented enhanced safeguards and revised pricing for the model. However, some experts warn that the release of such powerful AI could pose risks if not properly managed.
This launch marks a pivotal moment in the AI industry, as it brings cutting-edge capabilities that were once deemed too hazardous into the public domain.