Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, with a Claude Fable 5.1 pricing pitch aimed squarely at developers running costly multi-step AI agents. The company says Fable 5.1 should cost about 25% less than Fable 5 on typical token-billed workloads, and up to roughly 45% less on highly agentic work. That is an estimate tied to a particular mix of tokens, not a flat reduction on every completed task.
Fable 5.1 is generally available. Mythos 5.1 is limited to Anthropic’s trusted-access programs for cybersecurity and life-sciences work. Anthropic says the pair uses the same underlying model, with different safeguard levels.
How does Claude Fable 5.1 pricing reduce agent costs?
Anthropic cut the price of cache reads: inputs the model has already processed and stored. Agentic systems tend to revisit earlier material, such as a codebase, tool output, task history, or prior model responses, while they work through several steps. Cheaper reuse of that context can lower the bill where cached input makes up a substantial share of token use.
An LLM produces and consumes tokens rather than pulling a finished answer from a database. That billing detail is worth keeping straight when comparing models; LLMs work by processing and generating tokens, and an API bill depends on the mix of input, cached input, and output tokens a workload produces.
Anthropic’s 45% figure therefore does not establish that Fable 5.1 will be cheaper for every workload. The Decoder reported that Artificial Analysis found the lower cache-read rate saved about $1.40 per agentic task in its assessment, but that Fable 5.1 at maximum effort used roughly 1.7 times as many output tokens as Fable 5. In that analysis, the newer model cost 20% more per Intelligence Index task than Fable 5. That is a reported third-party assessment, not an Anthropic result, but it illustrates the basic nuisance hidden by a percentage headline: lower cached-input prices and lower total task cost are different measurements.
What else is Anthropic changing in Fable 5.1?
Anthropic says the new model improves coding, knowledge-work, and longer problem-solving performance compared with Fable 5. Those are company-reported evaluations, and Anthropic says production safeguards were enabled during testing. It also says safeguard interventions likely lowered scores on some benchmarks.
On safety controls, Anthropic says its cybersecurity safeguards generate 60% fewer false positives than before. The company says Fable 5.1 may help identify software vulnerabilities, while development of exploits remains prohibited. Claims about fewer erroneous blocks are still claims from the vendor, rather than an independent audit of real-world behavior.
Anthropic also announced Enterprise Frontier Safeguards, which it says will keep enterprise data in cloud infrastructure controlled by the customer rather than on Anthropic’s systems. The company says the program will reach enterprise customers in phases beginning later this fall. Until then, eligible customers can use Fable 5.1 with zero data retention, according to Anthropic.
For buyers, the useful comparison is cost per successful task under their own context reuse, effort setting, and output needs. Cache-heavy agents may see the savings Anthropic projects. Output-heavy maximum-effort jobs may not.
This story draws on original reporting from The Verge.
