Claude Fable 5.1 and Mythos 5.1: What Enterprises Need to Know

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on Tuesday, offering enterprises two versions of the same underlying model with different safeguards.

Fable 5.1 is generally available with Anthropic’s standard protections. Mythos 5.1 has more permissive safeguards for approved cybersecurity and life sciences work and is currently limited to certain US organizations through Anthropic’s trusted-access programs.

For enterprise customers, the practical difference is not just performance. Fable offers broad availability and lower cache-read costs, while Mythos is reserved for vetted organizations that require advanced cybersecurity or life sciences capabilities.

The pricing shift

The headline of this release isn't performance; it's economics. Anthropic slashed cache read pricing by 75%, from $1 to $0.25 per million tokens. 

Standard input and output pricing remains unchanged at $10 and $50 per million tokens, respectively. For typical workloads, Anthropic estimates costs drop by roughly 25%, with highly agentic tasks seeing savings up to 45%.

This matters because autonomous agents often reuse the same codebases, tool definitions, and conversation history. Without effective caching, agents may repeatedly incur standard input charges when resending code, tool definitions, and conversation history. Lower cache-read pricing could therefore make long-running, context-heavy tasks more economical.

The enterprise architecture

Anthropic is also introducing Enterprise Frontier Safeguards (EFS), a system that lets customers store monitoring data in their own cloud infrastructure rather than on Anthropic's systems. "EFS works by storing data in cloud infrastructure controlled entirely by the customer, not Anthropic," the company says. The rollout begins this fall; eligible customers can use Fable 5.1 with zero data retention until then. 

This addresses a persistent friction point for regulated industries: data sovereignty. EFS could make Fable more practical for organizations with strict privacy, monitoring, or data-retention requirements, though the system is not live yet. Anthropic says it developed EFS with more than 100 customers across industries including financial services, healthcare, manufacturing, telecommunications, law, retail, and the public sector.

Performance gains

In Anthropic’s testing, Fable 5.1 substantially outperformed Fable 5 across several coding, scientific research, and knowledge-work benchmarks. On Terminal-Bench-Science 0.1, it scores 52.6% versus Fable 5's 24.7%. On AutomationBench, it reaches 31.4% versus Fable 5's 17.1%. The model "sets a new standard for coding, knowledge work, and long-running problem-solving tasks," Anthropic says.

On Terminal-Bench 4.0, Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9%. Anthropic attributed the gap to tasks on which Fable’s safeguards intervened rather than a difference in the underlying model.

The models also produced some striking research results. Mythos 5.1 generated protein binders with a nearly 50% hit rate across 12 targets, while Fable 5.1 created a higher-resolution elevation map covering roughly one-third of Venus. Anthropic says external organizations experimentally validated the protein designs.

What eWeek found: Decoding the Fable vs. Mythos tradeoff 

The two models share the same underlying technology, but they are not interchangeable enterprise products. Fable is the broadly available option, while Mythos is a restricted configuration for approved cybersecurity and life sciences work. Here’s the breakdown:

Anthropic restricts edits to Claude’s prior reasoning

Anthropic also changed how new API accounts can handle Claude’s previous reasoning in multi-turn conversations. Those accounts can no longer manually alter earlier context while preserving the transcript of Claude’s prior thinking, closing what the company describes as a publicly documented model-distillation technique.

Existing accounts are not currently affected, although Anthropic says the restriction will apply to all users with future model releases and may affect a small number of custom integrations.

Limitations and risks

Anthropic reported that Mythos 5.1 improved on most alignment measures compared with Mythos 5. However, Anthropic's system card describes “a slight regression on overall misaligned behavior compared to Opus 5” and says Mythos 5.1 accepts unverifiable claims of authorization somewhat more readily than Opus 5.

External testing also identified a low-severity sandbox vulnerability that allowed Mythos 5.1 to access files outside its intended environment, according to the system card. 

The bottom line

For enterprises running context-heavy Claude agents, Fable 5.1’s lower cache-read price could produce meaningful savings, although the benefit will depend on how heavily each workload reuses stored context. Its broader cloud availability and more precise safeguards also make it the practical choice for most organizations.

Mythos 5.1 should be viewed as a restricted research configuration rather than a general enterprise upgrade. Enterprises should evaluate Fable against their own workloads and governance requirements, then watch whether EFS delivers the promised privacy and monitoring controls when it begins rolling out this fall.

Read more: Anthropic’s Claude Opus 5 offers another enterprise option for balancing model performance with operating costs.