Cache reads get 75% cheaper on Claude Fable 5.1

Anthropic has made Claude Fable 5.1 generally available, estimating that lower cache-read pricing will reduce typical token-billed workloads by about 25 percent, while reserving the identically capable Claude Mythos 5.1 for trusted cybersecurity and life-sciences access programs, and Anthropic says Fable 5.1 is available across its services, Amazon Web Services, Google Cloud, and Microsoft Azure.

At a glance

• The price change applies to cached context reads, inputs the model has already processed, and Anthropic puts their price at $0.25 per million tokens, 75 percent below the prior cache-read rate.
• Enterprise Frontier Safeguards will keep customer data in cloud infrastructure controlled by the customer, with default human review by that customer, rather than Anthropic, when the phased rollout begins later this fall.
• The generally available model can identify software vulnerabilities but not develop exploits, while Mythos 5.1 gives vetted cyberdefenders and life-sciences professionals safeguards designed for their restricted research and defensive work.

The release makes cost control, privacy requirements, and restricted research access parts of the same product decision. It appears aimed at organizations that need long-running coding agents but cannot treat retained prompts or blunt safety filters as incidental constraints. The separate Mythos channel also likely creates a way to offer more capable biology and cyber workflows without making those safeguards broadly available to ordinary accounts at this stage.

Fable 5.1 cuts cache-read pricing to $0.25 per million tokens

Fable 5.1 otherwise retains Fable 5 pricing of $10 per million input tokens and $50 per million output tokens wherever usage is token-billed. Anthropic says the new cache-read price is particularly consequential in context-heavy agentic workflows, where cached material accounts for much of a session’s bill and estimated savings can reach about 45 percent.

Anthropic’s prompting guidance says existing Fable 5 prompts should work without changes, but calls effort the main control for intelligence, latency, and cost. It recommends testing low, medium, high, xhigh, and max settings again because equal effort labels do not represent the same amount of thinking as on Fable 5.

Mythos 5.1 reached nearly a 50 percent binder hit rate across 12 targets

Anthropic tested Mythos 5.1 with open-source protein design and folding tools, then sent designs to two outside organizations for experimental validation. On three targets, it says, the resulting binding affinities were 10 times higher than the best designs submitted to Adaptyv Bio’s protein-design competitions. Across 12 targets, viable binders reached a nearly 50 percent hit rate.

Fable 5.1 also trained a neural network using Magellan radar images and an existing map to create a high-resolution elevation map for a third of Venus. Anthropic says the result resolves details at two to three kilometers instead of 10 to 20 kilometers, and estimates heights up to 25 percent more accurately. Separately, using custom GPU kernels and intermediate-result caching, Mythos 5.1 sped seven open-source models by up to 2.5 times with identical outputs.

Cyber safeguards produce around 60 percent fewer Claude Code interventions

Cybersecurity safeguards now allow Fable 5.1 to identify software vulnerabilities while continuing to redirect exploit generation, penetration testing, and binary-based vulnerability scanning to Opus models. Anthropic estimates that Claude Code users will encounter around 60 percent fewer cyber-safeguard interventions per session than with the prior Fable 5 safeguards.

Anthropic says its new Enterprise Frontier Safeguards were developed with more than 100 customers and cloud partners AWS, Google Cloud, and Microsoft Azure. The system is planned for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, and Microsoft Foundry.

Access remains restricted Mythos 5.1 is currently limited to a set of US organizations vetted for cyberdefense or life-sciences work. Anthropic says it is coordinating with the US government to widen access. The Cyber Verification Program is expected to add Mythos-class models, while the Life Sciences Verification Program plans broader enrollment. EFS begins its phased enterprise rollout later this fall.