Anthropic launches Claude Fable 5.1 with 75% cache cut

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, marking a significant update to its flagship AI model lineup just three months after launching the original Fable 5. The two models arrived on the first day of September 2026, with Fable 5.1 generally available now on the Claude API as claude-fable-5-1 and on Amazon Web Services, Google Cloud, and Microsoft Azure. The update brings enhanced coding capabilities and substantial cost reductions for developers using cache-intensive workflows.

Claude Fable 5.1 and Claude Mythos 5.1 logos launching September 1, 2026 with improved cache efficiency.

Two Models, Different Safeguards

The two names refer to the same underlying model. Fable 5.1 is the generally available version, with Anthropic’s production safeguards in place, while Mythos 5.1 is available through restricted-access programs for vetted cybersecurity and life-sciences organizations that need capabilities normally constrained by those safeguards. Claude Mythos 5.1 is offered separately, by invitation only, as part of Project Glasswing.

For the approved participants inside Project Glasswing, Anthropic has removed the safety classifiers and fallback restrictions that govern Fable 5.1, while Fable 5.1 retains safety classifiers alongside fallback options for high-risk tasks, keeping it suitable for broad commercial deployment.

Fable 5.1 vs Mythos 5.1 comparison: safety classifiers and guardrails vs minimal filters and unrestricted access.

Pricing and Cost Reductions

Per-token prices are unchanged from Fable 5 at $10 per million input and $50 per million output, but cache reads drop 75% to $0.25 per million tokens. This represents a substantial cost reduction for workflows that rely heavily on cached data.

Claude API token pricing comparison chart showing 75% savings with new $0.10 cache read price vs old $0.40 price and OpenA...

Anthropic estimates that cut alone makes typical workloads about 25% cheaper and highly agentic workloads up to about 45% cheaper. For comparison, OpenAI’s current promotional API pricing for GPT-5.6 Sol is $4 per million input tokens, $0.40 for cached input and $20 per million output tokens through at least Nov. 21.

Technical Capabilities and Benchmarks

Fable 5.1 and Mythos 5.1 each support one million tokens in context, with output capacity reaching 128,000 tokens per response. Claude Fable 5.1 is addressable on the Claude API as claude-fable-5-1, and Anthropic says it sets a new standard on coding, knowledge work, and long-running problem-solving.

On performance benchmarks, on AutomationBench, intended to measure business workflows, Fable 5.1 scores 31.4%, compared with 17.1% for Fable 5 and 26.9% for Opus 5, and on CursorBench 3.2.0, it reaches 73.4%. However, those numbers should be read as vendor-reported results rather than independent proof of superiority.

Claude Fable 5.1 and Claude Mythos 5.1 demonstrate the strongest overall cyber capabilities of any model Anthropic has released, and across its internal evaluation suite, they meet or exceed the cybersecurity performance of Claude Mythos 5.

Real-World Problem Solving

The standout highlight of Fable 5.1 is its refusal to take quick shortcuts when debugging software, as rather than applying band-aid fixes, the model hunts down the root causes of system failures. During real-world testing with investment firm Millennium, Fable 5.1 successfully identified the source of a rare internal system crash that had baffled human engineers and competing AI models for several years.

Root cause analysis infographic showing NullReferenceException debugging with Fable 5.1 error identification and resolutio...

Fable 5.1 adds two notable beta features that weren’t available in its predecessor, with the first being mid-conversation effort adjustment, letting users dial up or down how hard the model works on a given step without starting a new session.

Safeguards and Security

According to the official system card released alongside the models, Anthropic has determined that Mythos 5.1 does not cross the CB-2 threshold. The company applies multiple layers of protection, including real-time classifier guards, access controls for guard exemptions, a bug bounty program and threat intelligence, rapid-response options for jailbreaks, and security controls against model weight theft.

Concentric security layers diagram showing Real-Time Classifiers, Access Controls, Bug Bounty Program, Threat Intelligence...

As with Fable 5 and Opus 5, Anthropic reports no evidence of a critical-severity jailbreak. The company has also implemented new anti-distillation measures, with new API accounts created from September 1, 2026 onward no longer able to manually edit Claude’s prior context in a multi-turn conversation while preserving the transcript of Claude’s prior thinking.

Training Data and Knowledge

Both models feature a reliable knowledge cutoff of June 2026 and a training data cutoff of June 2026. This places the models among the most current in terms of training data as of their September 1 release date.

Key Facts

Claude Fable 5.1 and Claude Mythos 5.1 were released on September 1, 2026
Cache read costs dropped 75% to $0.25 per million tokens
Both models support one million tokens in context with 128,000 token output capacity
Fable 5.1 scores 31.4% on AutomationBench, compared with 17.1% for Fable 5
Millennium investment firm reported the model identified a system crash that had eluded engineers for years
Fable 5.1 is available on Claude API, AWS, Google Cloud, and Microsoft Azure