Anthropic Unleashes Claude Fable 5.1 and Mythos 5.1: Doubled Performance & 75% Cheaper Cache Reads

Anthropic Unleashes Claude Fable 5.1 and Mythos 5.1: Doubled Performance & 75% Cheaper Cache Reads

For readers tracking the shift, Anthropic continues its rapid innovation in artificial intelligence with the recent release of Claude Fable 5.1 and Claude Mythos 5.1. Arriving just three months after the Fable 5 line, these new large language models (LLMs) represent a significant leap forward, promising not only dramatic improvements in benchmark performance but also substantial cost efficiencies for developers and businesses.

Meanwhile, This latest update solidifies Anthropic’s position at the forefront of AI development, offering powerful tools for a wide range of applications, from complex scientific research to everyday agentic workflows.

Unveiling Fable 5.1 and Mythos 5.1: Key Distinctions

At their core, Fable 5.1 and Mythos 5.1 are built upon the same advanced underlying model. The primary distinction lies in their safeguard layers and availability:

• Claude Fable 5.1: This model is now generally available to the public via the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. It’s accessible to a broad spectrum of users and organizations.
• Claude Mythos 5.1: Reserved for specialized applications, Mythos 5.1 is restricted to vetted US organizations operating within Project Glasswing. This indicates its potential for highly sensitive or critical use cases where enhanced safeguards are paramount.

In practical terms, Both models boast an impressive 1 million token context window and a maximum output of 128,000 tokens, with Anthropic’s adaptive thinking capabilities always active, ensuring sophisticated reasoning across diverse tasks.

Dramatic Performance Leaps on Key Benchmarks

The headline-grabbing improvement comes in the form of significantly enhanced benchmark scores. Fable 5.1 demonstrates a remarkable doubling of performance in critical areas:

• Terminal-Bench-Science 0.1: Fable 5.1 achieves an impressive 52.6% on this agentic scientific research benchmark. This is a massive improvement compared to Fable 5’s 24.7% and even surpasses Opus 5’s 29.0% and GPT-5.6 Sol’s 22.4%.
• Terminal-Bench 4.0: Fable 5.1 scores 55.8%, while Mythos 5.1 reaches an even higher 60.9%. Anthropic transparently notes that the difference between these otherwise identical models reflects the “cost of safeguard interventions” in Mythos 5.1.

For example, Beyond these, the models show strong performance across other specialized benchmarks, including CursorBench 3.2.0 (73.4%), Humanity’s Last Exam (60.9% without tools, 65.0% with tools), AutomationBench (31.4%), OSWorld 2.0 (41.7% strict), and GDPval-AA v2 (1853).

Significant Cost Savings for Developers

Perhaps equally impactful for broad adoption is the substantial reduction in operational costs. Anthropic has slashed the price of cache reads by an impressive 75%, moving from $1.00 to just $0.25 per million tokens.

That said, While base input and output pricing remains unchanged, this cache read reduction translates to considerable savings:

• Approximately 25% lower overall cost for typical workloads.
• Up to 45% lower cost for context-heavy agentic workloads, where cache utilization is higher.

This makes the new Claude models a more economically attractive option for applications that frequently leverage cached information, enhancing efficiency without compromising performance.

Key API Changes for Developers to Note

Interestingly, Developers transitioning to Fable 5.1 should be aware of several breaking changes that could impact existing agentic setups:

• Forced Tool Use Removed: Setting tool_choice to any or tool will now result in a 400 error. Developers should instead use auto with strict tool use or structured outputs for tool invocation.
• Model-Bound Thinking Blocks: Thinking blocks are now specific to the model that generated them. This means that earlier models cannot read thinking from Fable 5.1, and vice-versa. Router and fallback configurations that switch between models may lose reasoning context.
• Editing Earlier Turns Invalidates Thinking: Modifying or injecting content into previous conversation turns, or rebuilding system/tools arrays mid-conversation, will now trigger errors. This check applies to accounts created on or after August 31, 2026. The recommended fixes involve using turn-scoped system messages or server-side context editing.

In addition to these breaking changes, Anthropic has introduced additive features (currently in beta behind headers) like per-message effort and turn-scoped system messages. Content provenance is also now mandatory, with a statistical text watermark on all output and C2PA credentials on files. Developers should also be aware of documented regressions, such as more variable parallel tool calling and a preference for whole-file rewrites over targeted edits.

Enhanced Safeguards and Real-World Scientific Impact

However, Anthropic has also refined its safety protocols, demonstrating a commitment to responsible AI development:

• Cyber Safeguards: Interventions in Claude Code for vulnerability discovery have been cut by roughly 60% per session, while still preventing exploit development. Penetration testing and exploit generation remain redirected to Opus.
• Biology Safeguards: These now fire 85% less often on benign requests, improving the user experience for legitimate biological research.

The scientific prowess of these models is already evident:

• Mythos 5.1 successfully designed protein binders with an approximate 50% hit rate across 12 targets, significantly outperforming the typical 10-15% norm.
• Fable 5.1 was used to build a Venus elevation map with 2-3 km resolution.
• Custom GPU kernels developed with these models sped up seven open-source genomics models by up to 2.5x.

Expert Perspective

From an industry angle, the clearest signal around Claude Fable 5.1 is how it may influence fable. The story reads less like a one-day spike and more like a marker of broader movement.

The next phase will depend on how quickly teams, regulators, or customers react. In practice, that gives Claude Fable 5.1 room to reshape expectations across claude over the near term.

For readers focused on practical impact, the best next step is to watch what changes around anthropic once attention turns into execution.

Frequently Asked Questions

Why does Claude Fable 5.1 matter right now?

Anthropic Unleashes Claude Fable 5.1 and Mythos 5.1: Doubled Performance & 75% Cheaper Cache Reads For readers tracking the shift, Anthropic continues its rapid innovation in artificial intelligence with the recent release of Claude Fable 5.1 and Claude Mythos 5.1.

What broader change could Claude Fable 5.1 signal?

Arriving just three months after the Fable 5 line, these new large language models (LLMs) represent a significant leap forward, promising not only dramatic improvements in benchmark performance but also substantial cost efficiencies for developers and businesses.

What should the market watch next around Claude Fable 5.1?

Meanwhile, This latest update solidifies Anthropic’s position at the forefront of AI development, offering powerful tools for a wide range of applications, from complex scientific research to everyday agentic workflows.

Conclusion

The headline is important, but the follow-through will shape the real outcome. Meanwhile, Claude Fable 5.1 and Mythos 5.1 represent a compelling advancement from Anthropic. With dramatically improved benchmark performance, substantial cost reductions for cache reads, and refined safety mechanisms, these models are poised to empower developers and researchers to achieve more. While developers will need to adapt to some breaking API changes, the overall benefits in terms of capability and efficiency make these new Claude iterations a powerful addition to the AI toolkit.

Source: https://www.marktechpost.com/2026/09/01/anthropic-releases-claude-fable-5-1-and-claude-mythos-5-1-52-6-on-terminal-bench-science-and-75-cheaper-cache-reads/