Claude Fable 5.1 and Claude Mythos 5.1 Release Notes

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, models designed for high-end coding, knowledge work, and scientific research. While both models share the same underlying architecture, they differ in their safeguard configurations: Fable 5.1 is the general-availability version, and Mythos 5.1 is restricted to trusted access programs for cybersecurity and life sciences professionals.

Cost Reductions and API Pricing

Claude Fable 5.1 introduces a significant price reduction for cache reads, reducing the cost by 75% to $0.25 per million tokens. This change specifically targets long-running agentic workflows where the model frequently re-reads processed context.

Typical Workloads: Estimated 25% overall cost reduction compared to Fable 5.
Agentic Workflows: Potential savings of up to 45% due to the high volume of cache reads.
Standard Pricing: Input tokens remain at $10 per million and output tokens at $50 per million.

Scientific Research and Computational Capabilities

Claude Mythos 5.1 demonstrates advanced capabilities in molecular design and computational biology, often outperforming human specialists or existing open-source tools.

Molecular Design and Biology

In protein design, Mythos 5.1 achieved a hit rate of nearly 50% across 12 targets when designing high-affinity binders, significantly higher than the typical 10-15% hit rate in current protein design practices. Additionally, the model optimized seven open-source deep learning models by writing custom GPU kernels, resulting in inference speed-ups of up to 2.5x and GPU cost reductions of 30-60%.

Planetary Science

Claude Fable 5.1 was used to create a high-resolution elevation map of one-third of the planet Venus, utilizing 30-year-old NASA Magellan radar images. The resulting map provides details down to 2-3 kilometers (compared to the previous 10-20 kilometers) and improves height accuracy by up to 25%.

Performance Benchmarks

Fable 5.1 shows marked improvements over Fable 5 and other frontier models in agentic scientific research and coding.

Benchmark	Fable 5.1	Fable 5	Opus 5	GPT-5.6 Sol
Terminal-Bench-Science 0.1	52.6%	24.7%	29.0%	22.4%
Agentic Coding (Terminal-Bench 4.0)	55.8%	42.0%	52.3%	37.3%
Knowledge Work (GDPval-AA v2)	1853	1723	1824	1711
OSWorld 2.0 (Partial)	77.9%	72.9%	75.4%	—
Humanity's Last Exam (No Tools)	60.9%	57.8%	56.6%	—
AutomationBench	31.4%	17.1%	26.9%	19.6%

Note: Mythos 5.1 scored 60.9% on Terminal-Bench 4.0.

Privacy and Enterprise Safeguards

Anthropic is introducing Enterprise Frontier Safeguards (EFS), a system that provides the privacy of a zero data retention policy while maintaining the ability to detect adversarial use. EFS stores data in cloud infrastructure controlled entirely by the customer rather than Anthropic.

Availability: Rolling out in phases starting fall 2026 across AWS, Google Cloud, and Microsoft Azure.
Interim Solution: Eligible customers can use Fable 5.1 with zero data retention until EFS is fully deployed.

Security and Alignment

Cybersecurity and Biology Safeguards

Fable 5.1 reduces false positives in cybersecurity tasks by 60% and is now permitted to identify software vulnerabilities for defensive purposes. However, it still redirects dual-use tasks—such as exploit generation and penetration testing—to Opus models. In biology, safeguards for benign requests (medical/elementary biology) fire 85% less often than in Fable 5.

Anti-Distillation and Watermarking

To prevent the illicit extraction of model capabilities, Fable 5.1 implements strengthened anti-distillation mechanisms. New API accounts can no longer manually edit prior context in multi-turn conversations while preserving the model's thinking transcript.

Additionally, in compliance with the EU AI Act, Anthropic has implemented invisible watermarking for text outputs. A detection API is currently in private preview for regulators, law enforcement, and eligible organizations.

Community Insights and Counterpoints

Users and developers on Hacker News provided a mix of qualitative feedback and technical critiques:

Writing Style: Some users noted a more natural prose style with fewer stock phrases, while others argued the output has become denser and less succinct.
** Endgame for Frontier Progress:** Some critics questioned if the marginal gains in benchmarks (excluding science) suggest a plateau in frontier model progress.
Cost vs. Value: Despite the cache discount, some developers argued that the cost remains prohibitively high for general coding agents compared to cheaper alternatives like DeepSeek or GLM.
Technical Quirks: Users reported that Fable 5.1 is more prone to rewriting entire files for small changes, which increases token consumption. Anthropic recommends adding a specific system prompt instruction to encourage surgical edits.

"In internal benchmarks, Claude Fable 5.1 solves more of our coding problems than Fable 5 or Opus 5, and achieves state of the art on trading intuition." — Craig Falls, Head of Quantitative Research, Jane Street Capital