Claude Fable 5.1 Pricing: API Rates, Plans, and Budget Example

On September 1, 2026, Anthropic launched Claude Fable 5.1, a new AI model built for advanced coding and knowledge work, with additional safeguards for enterprise and scientific use. Claude Fable 5.1 and the more restricted Mythos 5.1 share the same underlying architecture, but differ in their access levels and security controls.

Compared with earlier Anthropic models, including Fable 5, Claude Fable 5.1 offers stronger performance, better safeguards, and significantly lower API pricing for many workloads. It delivers similar or better results at roughly 25% lower cost for typical token-based workloads, partly because of cheaper cache reads. For complex, agentic tasks, costs can be up to 45% lower than before.

These changes make high-end AI more practical and affordable for regulated businesses and teams working in law, finance, healthcare, and compliance-focused operations. Better privacy, lower costs, and more refined safeguards could shape how these organizations approach LLM integration, particularly in workflows that require strong data security, auditability, and predictable spending.

API Pricing: Per-Token Rates for Claude Fable 5.1

Claude Fable 5.1 uses a per-token billing model for API access, with separate rates for input and output tokens depending on the task type and effort setting. Anthropic lists Fable 5.1 at $10 per million input tokens and $50 per million output tokens — the same base rates as Fable 5. What changed is the cache-read rate, which falls from $1 to $0.25 per million tokens.

The reduction is mainly due to Anthropic’s decision to lower prices on cache reads, which affects workloads that reuse previously processed data. That single change is the whole of the advertised saving: Anthropic prices a Fable 5.1 cache hit at 0.025x the base input rate, where every other Claude model charges 0.1x. Anthropic puts the effect at about 25% less for typical workloads and up to roughly 45% for highly agentic work, because those runs re-read a large cached context on every step.

The full rate card is published on Anthropic’s pricing page: $10 input, $50 output, $0.25 cache read, and cache writes at $12.50 for the 5-minute window or $20 for the 1-hour window, all per million tokens. Batch processing halves input and output to $5 and $25. Rates change, so confirm them before you budget.

Included Plans: Team, Enterprise, and Promotion Offers

Claude Fable 5.1 is generally available through Anthropic’s API and products, but specific plan inclusions are governed by Anthropic’s current offering structure. The model is accessible to most customers through team and enterprise plans; Claude Mythos 5.1 remains restricted to those in approved access programs, particularly in cybersecurity and the life sciences.

There is no mention of a fixed introductory or promotional rate for Claude Fable 5.1 in the announcement. Promotional access is subject to Anthropic’s phases of rollout and eligibility criteria, and may be updated as enterprise features and safeguards (like Enterprise Frontier Safeguards) become more widely available later in 2026.

Small teams and consumer accounts are likely to have access to Claude Fable 5.1 through the main API endpoints, but should confirm their eligibility and rates with Anthropic’s documentation as terms may differ by tier.

How Claude Fable 5.1 Pricing Compares to Fable 5 and Flagships

Claude Fable 5.1 costs exactly what Fable 5 costs on fresh input and on output: $10 and $50 per million tokens. The saving is real but conditional — it lands only on cached input, at $0.25 against Fable 5’s $1. A workload that never reuses context pays what it paid before; one that re-reads a large cached prompt on every step sees the full benefit.

Relative to the main competitor flagships such as GPT-5.6 Sol, Claude Fable 5.1 now occupies a more competitive position for both performance and price in coding and knowledge-heavy business workloads. Public benchmarks cited by Anthropic show Fable 5.1 outperforming Fable 5 and GPT-5.6 Sol on several agentic and scientific reasoning tasks.

Against the current OpenAI flagship, GPT-5.6 Sol at $5 input and $30 output per million tokens, Fable 5.1 is the more expensive model on raw rates. Its case rests on cached input, where $0.25 undercuts Sol’s $0.50. Which model is cheaper for you depends on how much of your input arrives as cache hits, so price both against your own token mix rather than against either vendor’s headline.

• 25% lower effective cost vs Fable 5 for typical workflows
• Up to 45% lower for agentic/scientific research scenarios
• Positioned competitively with GPT-5.6 Sol and prior Claude flagships

Example: Budgeting a Realistic Monthly Claude Fable 5.1 Bill

While Anthropic does not provide exact figures in the announcement, a typical monthly bill depends on the input and output token counts, model effort settings, and rate of cache-eligible reads. For a mid-sized firm running document analysis and complex coding tasks, token usage can easily reach 10 to 50 million tokens per month depending on workflow scale.

Take a team spending $5,000 a month on Fable 5. If none of that input is cached, the bill does not move, because base rates are unchanged. If the workload matches what Anthropic calls typical, the saving is about 25%, or roughly $3,750. A heavily agentic workload that re-reads a large cached context on every step approaches 45%, or about $2,750. The figure depends almost entirely on what share of input tokens are cache hits, which is worth measuring before you forecast.

Every buyer should model their own expected costs by measuring token counts across input and output, and by reviewing up-to-date rates on Anthropic’s official pricing page before making a commitment.

Who Should Use Fable 5.1 Pricing vs Smaller Models

The new Claude Fable 5.1 pricing best serves teams and businesses running complex knowledge work, agentic research, or software engineering with large context requirements, especially those in regulated industries that require strong safeguards and compliance features.

Buyers with high-volume, low-complexity language tasks—such as basic summarization, content generation, or repetitive data extraction—can often reduce costs by routing those requests to smaller, less expensive models where maximal safety or capability is not required.

A careful split of task types lets firms control costs and quality at once: use Fable 5.1 for workloads that need peak coding, reasoning, or privacy, and supplement with lighter models for commodity text processing.