Anthropic Launches Claude Fable 5.1 and Mythos 5.1, Doubling Programming Benchmarks and Slashing Cache Read Costs by 75%

On September 1st, Anthropic unveiled Claude Fable 5.1 and Claude Mythos 5.1, positioning them as the world’s most advanced models for programming and knowledge work. Built on a shared base architecture, the models differ in their security protocols: Fable 5.1 is accessible to all users, whereas Mythos 5.1 is restricted to vetted organizations in cybersecurity and life sciences via the Trusted Access Program.

Benchmark Performance Surpasses Predecessors at Lower Costs

Benchmark results show Fable 5.1 outperforming its predecessor, Opus 5, and OpenAI’s GPT-5.6 Sol across multiple metrics. In the Agent Science Research Benchmark (Terminal-Bench-Science 0.1), Fable 5.1 achieved 52.6%, significantly higher than Fable 5 (24.7%), Opus 5 (29.0%), and GPT-5.6 Sol (22.4%). For the Agent Programming Benchmark (Terminal-Bench 4.0), Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9% under relaxed security settings. In knowledge work tasks, Fable 5.1 scored 1853 on the GDPval-AA v2 test, surpassing Opus 5’s 1824 and Fable 5’s 1723.

Regarding pricing, cache read costs for Fable 5.1 have dropped by 75% to $0.25 per million tokens. Under typical workloads, costs are approximately 25% lower than Fable 5, and up to 45% lower for highly agent-driven tasks. Base pricing remains unchanged at $10 per million input tokens and $50 per million output tokens.

Anthropic has enhanced Fable 5.1’s security mechanisms to minimize false positives, reducing cybersecurity-related false interceptions by roughly 60% compared to previous versions. While Fable 5.1 can now assist in identifying software vulnerabilities, it is restricted from developing exploits for them. In the biological sector, Anthropic has collaborated with the U.S. government to create an access program for researchers using Mythos 5.1, with public registration for scientists planned soon.

Scientific Breakthroughs, Enhanced Compliance, and Data Privacy

Anthropic highlighted Mythos 5.1’s capabilities across several scientific domains. In molecular design, the model used open-source protein folding tools to create high-affinity binders, achieving binding affinities ten times higher than top entries in the Adaptyv Bio Protein Design Competition for three targets. It also achieved hit rates near 50% on 12 targets, outperforming the typical 10–15% hit rate in protein design. In computational analysis, Fable 5.1 trained a neural network using NASA’s Magellan mission radar data from over 30 years ago, generating a high-resolution elevation map covering one-third of Venus. This map refined resolution from 10–20 kilometers to 2–3 kilometers, improving elevation accuracy by 25%. In computational biology, Mythos 5.1 accelerated seven open-source deep learning models by up to 2.5 times through custom GPU kernels and intermediate result caching, potentially reducing GPU costs by 30–60%.

For data retention, Anthropic introduced Enterprise-Focused Security (EFS), combining zero-data-retention privacy with robust security measures. Customer data is stored within cloud infrastructure fully controlled by the enterprise, not Anthropic’s systems. EFS will be gradually rolled out to enterprise customers this fall; until then, eligible users can access Fable 5.1 under the zero-data-retention policy.

Regarding compliance, Anthropic adopted the EU AI Act’s transparency guidelines for AI-generated content in July 2026. Under this regulation, model outputs released after August 2, 2026, must include an invisible watermark. Fable 5.1 now embeds this watermark in generated text, which remains undetectable to users without the detection API, does not impact output quality, and contains no user or conversation data. Anthropic is launching the detection API in private preview, allowing qualified entities such as regulatory bodies, law enforcement, media, and fact-checking organizations to verify content in accordance with EU laws.

Claude Fable 5.1 is now available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, accessible via the Claude API. Claude Mythos 5.1 remains available exclusively to approved U.S. organizations.