Anthropic Debuts Claude Fable 5.1 and Mythos 5.1 with Aggressive Caching Discounts

Anthropic has rolled out major upgrades to its frontier AI roster, making Claude Fable 5.1 generally available alongside a restricted release of Claude Mythos 5.1, as announced in Introducing Claude Fable 5.1 and Claude Mythos 5.1. While standard API token rates remain steady at $10 per million input tokens and $50 per million output tokens, the company has slashed the cost of prompt cache reads by 75 percent—dropping them from $1.00 down to $0.25 per million tokens according to Claude Fable 5.1 - Claude Platform Docs. Alongside this pricing adjustment, Anthropic posted self-reported benchmark scores that more than double its previous generation on complex, autonomous scientific benchmarks, sharpening the competitive race against OpenAI and Google to capture enterprise agent workloads.

Background

Over the past year, the artificial intelligence industry has pivoted from conversational chatbots to autonomous software agents—systems engineered to plan, execute code in terminal environments, and solve multi-step problems over extended periods. However, deploying multi-turn agents poses two primary hurdles for enterprises: cost and safety.

Because agents continually pass lengthy system instructions, tool definitions, and ongoing conversational history back into a model, repeating this context can rapidly inflate token usage. To address this overhead, providers developed prompt caching, storing previously processed text in memory so the model can reuse context rather than re-evaluating it from scratch. As enterprise adoption matures, pricing on these repeated "cache reads" has emerged as a crucial lever in lowering total cost of ownership for long-running workflows.

What Happened

Anthropic officially debuted Claude Fable 5.1 and Claude Mythos 5.1, two variations built on the same underlying model architecture but separated by guardrails and distribution models, as outlined in Coursiv Blog's overview.

Claude Fable 5.1 is now generally available via the Claude API, Claude Enterprise, and major cloud partners. It features a 1-million-token context window and an output capacity of up to 128,000 tokens. While base input and output rates are unchanged, the 75 percent reduction in cache read pricing—down to $0.25 per million tokens—lowers the blended operational cost by an estimated 25 percent for typical workloads and up to 45 percent for context-heavy agentic tasks, according to byteiota.

Anthropic also documented significant performance jumps on agent-oriented evaluations reported by Gadgets 360:

• On Terminal-Bench-Science 0.1, which evaluates end-to-end command-line scientific tasks, Fable 5.1 reached 52.6 percent, up from 24.7 percent on Fable 5.
• On Terminal-Bench 4.0, it scored 55.8 percent, compared to 42.0 percent previously.
• On AutomationBench, scores rose to 31.4 percent from 17.1 percent.

In contrast to Fable 5.1's public availability, Claude Mythos 5.1 is restricted to vetted life scientists and cybersecurity defenders under trusted access initiatives. It relaxes certain production safeguards to support sensitive research without tripping false alarms on benign queries. Anthropic additionally previewed its upcoming Enterprise Frontier Safeguards (EFS), designed to store client data within customer-controlled cloud infrastructure to achieve zero-retention compliance.

Why It Matters

The dual announcement signals a strategic pivot in frontier AI economics. By holding baseline rates steady while cutting cache read fees, Anthropic is directly targeting the recurring costs associated with agentic systems. Rather than competing purely on base token prices, the company is using prompt caching discounts to make context-heavy enterprise deployments considerably cheaper.

Furthermore, the split-release structure between Fable and Mythos addresses the challenge of model alignment. Frontier models often risk "over-refusal," declining legitimate research requests in sensitive domains such as biological analysis or network defense. By creating a specialized pipeline for verified organizations, Anthropic provides advanced tooling to researchers while maintaining broad safeguards for general enterprise customers.

What's Next

In the coming months, attention will center on the phased rollout of Anthropic's Enterprise Frontier Safeguards (EFS) and whether access to Mythos 5.1 will expand beyond vetted U.S. entities to international partners. Industry watchers will also monitor independent evaluations to verify how Anthropic's self-reported benchmark gains hold up in production environments.

As rivals OpenAI and Google refine their own developer platforms, the focus will increasingly turn toward compute-effort controls and cache economics. If developers find they can achieve superior multi-turn agent performance at lower sustained costs, Anthropic's latest release could set a new baseline for enterprise agent architectures.