DEVOURED
Claude Fable 5 Launch
Anthropic launched Claude Fable 5, a powerful model for general use, alongside Claude Mythos 5, an unrestricted version for high-stakes cybersecurity and research.
What: Claude Fable 5 and Mythos 5 are available at $10/million input tokens and $50/million output tokens. Fable 5 includes safety classifiers that default to Opus 4.8 for sensitive cyber or biology queries, while Mythos 5 removes these constraints for vetted partners.
Why it matters: The tiered release strategy, with automatic fallback models for sensitive tasks, indicates how AI companies are attempting to commoditize frontier reasoning while containing dual-use risks.
Deep dive
- Fable 5 is state-of-the-art on benchmarks including FrontierCode and Hebbia’s Finance Benchmark.
- Mythos 5 enables autonomous scientific research, including protein design and genomics, with performance matching human experts.
- Introduces a 30-day data retention policy for business customers to improve safety and jailbreak defense.
- Fable 5 provides a mechanism to fallback to Claude Opus 4.8 when safety classifiers trigger.
- Uses persistent file-based memory to boost performance on long-horizon tasks.
Decoder
- Dual-use: Technologies that have both legitimate beneficial applications and significant potential for harm (e.g., bio-research tools that could also be used to design pathogens).
- Jailbreak: Techniques or prompts designed to circumvent an AI model's built-in safety guardrails.
Original article
Full article content is not available for inline reading.
Explore all 52 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi