Anthropic Releases Claude Fable 5.1 and Mythos 5.1, Cuts Agentic Costs by Up to 45%

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, updating its flagship model line with sharper agentic coding performance and lower costs for high-volume workloads. Both models share the same base model but differ in safety guardrails, following the pattern set by Fable 5 and its predecessors.

Fable 5.1 is broadly available now via API under the identifier claude-fable-5-1, on Anthropic's own platform and through AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 is restricted to two vetted programs in the US: the Cyber Verification Program for defensive security work and the Life Sciences Verification Program, developed with the US government. Anthropic says it plans to expand Mythos access to international partners over time.

Pricing cut targets agentic workloads

Standard API pricing for Fable 5.1 is unchanged at $10 per million input tokens and $50 per million output tokens — twice what Anthropic charges for Opus 5 ($5/$25 per million). The change comes in cache-read pricing, cut from $1 to $0.25 per million tokens. Anthropic says this reduces costs by roughly 25 percent on typical tasks and up to 45 percent on heavily agentic workflows involving long autonomous runs with many tool calls.

The cut addresses what Anthropic itself acknowledges was the biggest complaint about Fable 5: cost. High pricing likely contributed to weak enterprise adoption, a problem that intensified after Opus 5 launched in late July already matching or beating Fable 5 on most benchmarks at half the price.

Benchmark gains, according to Anthropic

Anthropic reports Fable 5.1 scoring 52.6 percent on Terminal-Bench-Science 0.1, more than double Fable 5's 24.7 percent and ahead of GPT-5.6 Sol's 22.4 percent and Opus 5's 29.0 percent. On Terminal-Bench 4.0 for agentic coding, Fable 5.1 scores 55.8 percent and Mythos 5.1 reaches 60.9 percent, compared to 42.0 percent for Fable 5, 52.3 percent for Opus 5, and 37.3 percent for GPT-5.6 Sol.

Other reported figures: GDPval-AA v2 at 1853 (vs. 1723 for Fable 5, 1824 for Opus 5, 1711 for GPT-5.6 Sol); OSWorld 2.0 partial-credit computer-use score at 77.9 percent; Humanity's Last Exam at 60.9 percent without tools and 65.0 percent with tools; AutomationBench at 31.4 percent; and CursorBench 3.2.0 at 73.4 percent. Anthropic says Fable 5.1 tops third-party benchmarker Artificial Analysis's Intelligence Index with a score of 66, ahead of Opus 5 (63) and GPT-5.6 Sol (61). These are Anthropic-reported and third-party aggregator figures; independent verification across real-world tasks is still pending.

Anthropic researcher Felix Rieseberg also claims improved writing style, saying Fable 5.1 relies less on bullet points and bold text in chat and follows style instructions more closely.

Watermarking and safety changes

Fable 5.1 and Mythos 5.1 are the first Claude models to ship with built-in watermarks. Anthropic is launching a detection API in private preview, letting regulators, media outlets, fact-checkers, and research institutions verify whether a given text contains the watermark, with plans to widen access over time.

Safety filters for cybersecurity, biology, and medical queries have been loosened after Fable 5's filters produced frequent false positives on well-intentioned requests. Anthropic claims the cybersecurity filters now generate 60 percent fewer false positives and biology-related filters fire 85 percent less often on harmless questions. Fable 5.1 can now identify software vulnerabilities, though not develop exploits — penetration testing and exploit generation remain routed to Opus models.

Anthropic is also closing a distillation-attack vector: new API accounts can no longer edit Claude's prior context in multi-turn conversations while retaining the thinking transcript, a technique the company says was used to systematically extract model capabilities through large numbers of fake accounts.

Enterprise customers get access to new Enterprise Frontier Safeguards (EFS), which keep customer data solely on the customer's own cloud infrastructure.

What this means

The pricing move is a direct response to competitive pressure. Once Opus 5 undercut Fable 5 on both price and benchmarks, Anthropic's flagship line needed a cost fix, not just a capability bump — hence targeting cache reads, the line item that dominates spend in long agentic sessions, rather than base token rates. Whether the benchmark gains hold up in production use, particularly against GPT-5.6 Sol on agentic coding, will depend on real-world deployment over the coming weeks rather than the benchmark tables Anthropic has published. The watermarking rollout and distillation crackdown also signal Anthropic is increasingly building provenance and anti-extraction infrastructure directly into model releases rather than treating them as separate policy layers.