Skip to content
NewsModel release

Anthropic Releases Fable 5.1 and Mythos 5.1 With 75% Cache-Read Price Cut and Fewer Security False Positives

· by Pondero Newsdesk

The short version

Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1, cutting cache-read prices 75% to $0.25 per million tokens and reducing cybersecurity false positives by 60%.

Anthropic Releases Fable 5.1 and Mythos 5.1 With 75% Cache-Read Price Cut and Fewer Security False Positives

Cache reads on Fable 5.1 now cost $0.25 per million tokens, down 75% from what Fable 5 charged, cutting typical production costs by around 25% and agentic workloads by up to 45%. Input and output token pricing did not change.

What happened

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. The two models share the same underlying weights but operate under different safeguard profiles, per the Anthropic announcement. Fable 5.1 is generally available through the API, Amazon Bedrock, Google Cloud, and Microsoft Azure under the model identifier claude-fable-5-1. Mythos 5.1 remains restricted to vetted U.S. organizations doing cybersecurity research through the Cyber Vulnerability Program or life sciences work through a new Life Sciences Verification Program, which Anthropic developed in partnership with the U.S. government.

The headline pricing change: cache reads drop to $0.25 per million tokens, a 75% reduction. Input and output prices hold at $10 and $50 per million tokens respectively. Anthropic measured the cost impact against real August 2026 usage data. Typical workloads save roughly 25% versus Fable 5. Context-heavy, tool-heavy agentic tasks save around 45%, because those sessions are disproportionately made up of cache reads rather than fresh input or output tokens.

On safeguards, Anthropic redesigned its cybersecurity filters. The new filters block 60% fewer false positives than the previous design. Fable 5.1 can now be used to discover software vulnerabilities, though not to develop exploits. Invisible text watermarking ships with both models, and Anthropic launched a detection API in private preview for organizations required to use it under EU law.

Enterprise data handling shifts as well. Anthropic announced Enterprise Frontier Safeguards (EFS), which stores customer activity data in cloud infrastructure the customer controls rather than on Anthropic's own infrastructure. EFS gives enterprise customers the privacy equivalent of a zero-data-retention policy while still allowing cross-session misuse detection. The rollout begins later in fall 2026 in phases. Until EFS is available, eligible customers can use Fable 5.1 with zero data retention, per Anthropic.

Why it matters

For any team running Fable 5 in production, the cache-read cut is a price change they can act on today without any code change. Agentic pipelines and long-context Claude Code sessions already carry a large share of their cost in cache reads, which is precisely why Anthropic's own measurement shows up to 45% savings on those workloads. Teams that have been modeling budgets assuming Fable-class pricing for production agents can revise those numbers downward now.

The false-positive reduction on cybersecurity tasks opens Fable 5.1 to uses that were previously routed to smaller models. Security teams doing penetration testing, bug bounty work, or internal vulnerability scanning had hit filter walls that blocked legitimate tool calls on Fable-class models. The 60% drop in false positives addresses that directly. Fable 5.1 can now assist with vulnerability discovery, a category Anthropic had previously blocked entirely at this safeguard tier.

EFS solves an enterprise adoption problem that has kept regulated-industry teams on lower-tier models. Anthropic's 30-day data retention requirement for Fable-class models conflicted with data-residency rules at banks, healthcare organizations, and other regulated operators. EFS lets those teams take Fable 5.1 into production under their own audit controls and encryption keys, which is what compliance teams in those industries require.

Context and reactions

Cognition, the maker of the Devin software engineering agent, said it would move Opus 5 traffic to Fable 5.1 on launch day. In a quote included in the Anthropic announcement, co-founder and CPO Walden Yan said the new cache-read pricing makes "a Fable-class model finally economical for the workloads we'd kept on Opus, starting with code review." Investment firm Millennium provided a testimonial describing Fable 5.1 finding the root cause of a rare software crash that had gone unexplained for several years.

On benchmarks, Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, a new agentic scientific research benchmark, against 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol, per the Anthropic page. On Terminal-Bench 4.0 for agentic coding, Fable 5.1 reached 55.8% and Mythos 5.1 reached 60.9%, with production safeguards active. Fable 5 scored 42.0% and GPT-5.6 Sol scored 37.3% on the same benchmark. On AutomationBench for business workflows, Fable 5.1 scored 31.4% against 17.1% for Fable 5. On CursorBench 3.2.0, Fable 5.1 scored 73.4% versus 70.5% for Fable 5.

TechCrunch reported that the system card rates Mythos 5.1 as "low-risk" for automated AI development, meaning it judges the risk of the model accelerating uncontrolled self-improvement to be in line with current trends, per TechCrunch. The same card acknowledges Mythos 5.1 is a "slight regression on overall misaligned behavior compared to Opus 5," cooperating with human misuse "somewhat more readily than Opus 5" while also being less likely to hallucinate or falsely claim task completion.

What to watch next

The EFS rollout timeline is the near-term milestone. Anthropic committed to "phases, starting later this fall" but has not published eligibility criteria or specific dates. The Life Sciences Verification Program enrollment is described as opening "soon" with no date given. On the competitive side, the Fable 5.1 cache-read cut puts pressure on OpenAI's caching pricing for GPT-5.6 and Codex; a pricing response from OpenAI would be the signal to watch.

Sources