Claude Fable 5.1 (2026): What It Is, What It Costs, Which Plans Include It, and When to Use Opus 5 Instead

Claude Fable 5.1: the short answer

Every launch week, someone on LinkedIn tells you to switch everything to the new flagship. Anthropic's own documentation says the opposite. As of September 2026, Claude Fable 5.1 is Anthropic's most advanced generally available model, announced September 1, 2026 alongside Claude Mythos 5.1. On the API it costs $10 per million input tokens and $50 per million output: twice Claude Opus 5 and five times Sonnet 5. It is not on the Free or Team plans, Enterprise includes it, and Pro and Max meter it separately. Anthropic's model guide says to start with Opus 5 for most workloads and move to Fable when Opus still falls short. This page is what Fable is, the full price sheet (including the caching rate that changes the math), which plans get it, and how a business should decide, with every number read off Anthropic's pricing page, plan table, and model docs on September 13, 2026.

What Claude Fable 5.1 is

Fable is the tier Anthropic added above Opus with the Claude 5 generation. Fable 5 came first; Fable 5.1 and Mythos 5.1 were announced together on September 1, 2026, in a post Anthropic's newsroom headlines as "our most advanced models for coding and knowledge work." Mythos 5.1 is listed at the same price as Fable 5.1 with "limited availability," so for a business buyer, Fable is the top of the menu you can actually order from.

Anthropic's model overview positions it narrowly: use Claude Fable 5.1 "for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5 at higher effort still fall short." The same comparison table rates its latency "Slower," the slowest of the four current models, and its API model ID is claude-fable-5-1. Like every Claude model from 4.6 on, it gets the full 1M-token context window at standard pricing, per Anthropic's pricing page: a 900K-token request bills at the same per-token rate as a 9K one.

Read that positioning twice. The company selling the model is telling you it is the escalation, not the default. The full ranking by business job is on best Claude model for business.

Claude Fable pricing

From Anthropic's pricing page, per million tokens, as of September 13, 2026:

Two things on that sheet matter more than the headline. First, the output rate: Fable's $50 per million output tokens is where the bill lives, because agentic work writes a lot. Second, the cache rate. Anthropic prices Fable 5.1 cache hits at 0.025x the base input price, where every other model uses 0.1x. That makes a Fable 5.1 cache hit $0.25 per million tokens, half the price of an Opus 5 cache hit and a quarter of what Fable 5 charges ($1). Writing to the cache still costs $12.50 per million for five minutes or $20 for an hour. If your workload re-reads the same long context, most of Fable's input premium disappears. What caching and batch do to the rest of the lineup is on Anthropic API pricing.

The math on a real workload

Take an agent that runs 1,000 times a month. Each run re-reads the same 50,000-token context (policies, product docs, a map of the codebase), adds 2,000 tokens of new input, and writes 1,500 tokens. Before cache writes:

• Fable 5.1: 50M cached tokens at $0.25 is $12.50, plus 2M input at $10 is $20, plus 1.5M output at $50 is $75. About $108 a month.
• Opus 5: $25 cached, $10 input, $37.50 output. About $73.
• Sonnet 5: $10 cached, $4 input, $15 output. About $29.

Without caching, the same Fable workload is 52M input tokens at $10 plus the output: about $595, against about $298 on Opus 5. So caching moves Fable from 2x Opus to roughly 1.5x on this job. The honest takeaway: on a cached, output-light workload the gap is small money for a business. On an output-heavy one you are paying double, and that is exactly where you should prove Opus 5 is failing before you pay it.

Which Claude plans include Fable

Anthropic's plan comparison on claude.com/pricing has a Fable row in both tables. Here is what it says, verbatim where it matters:

In plain terms: on Pro, the table lists usage credits rather than included usage, so budget Fable as extra spend; on Max, Fable can take up to half of your weekly limits, with a footnote worth reading on the page; Team does not include it at all; Enterprise does. That Team line is the one most small businesses miss. Team is the plan most owner-led companies buy, and the model in every launch-week headline is not on it. The rest of the plan-by-plan picture is on Claude AI pricing.

Fable vs Opus 5 vs Sonnet 5: how a business should decide

• Sonnet 5 for daily team work. Drafting, summarizing, email, and everyday analysis. At $2 / $10 on the API and the lightest draw on a subscription's limits, it is the default I set for clients.
• Opus 5 for the hard jobs and agentic coding. Anthropic's own starting recommendation for most API workloads, at 2.5x Sonnet 5 per token.
• Fable 5.1 as the escalation. Long-running agents, the analysis where a wrong answer costs real money, and anything you have already run on Opus 5 at higher effort and watched fail. Small volume, high stakes.

The failure I see is the reverse: a team switches its default to the newest model, burns through its limits by Wednesday, and decides "Claude got worse." The model dropdown is a budget decision on every plan, even the flat-priced ones.

Claude Fable vs GPT-6 Astra

On list price they are twins: OpenAI's GPT-6 Astra is also $10 in and $50 out per million tokens. The differences are in the fine print. Astra has a line at 272,000 input tokens past which the whole request reprices, while Anthropic bills Fable's full 1M-token window at the standard rate, and Fable 5.1's cache hits cost a quarter of the usual cache rate, which rewards workloads that re-read the same context. If you are choosing a top-tier model for a business agent, the pricing structure is the tiebreaker, not the benchmark chart. The OpenAI side is broken down on GPT-6 Astra.

Should a small business use Claude Fable?

Mostly, not yet. If you are on Claude Team, you do not have it, and you do not need it for the work a Team plan is for. If you are on Max, it is a lever for the handful of deep tasks a month that justify spending weekly allowance. If you are building on the API, run your hardest real task on Opus 5 at higher effort first, and move only the steps that still fail. The expensive mistake is not paying $50 per million output tokens. It is paying it for email.

What to do today

Find the one task Claude got wrong for you last month: the analysis that missed something, the agent run that wandered off. That is your Fable test. Run it once on Opus 5 with more effort, then once on Fable if your plan or API account has it, and compare the output, not the vibes. Leave your team's default model alone. And if you want Claude set up properly for the whole company, with the right model on the right job instead of everyone guessing at the dropdown, that is the Claude for small business install, from $4,500 in about two weeks; the $397 Playbook is the do-it-yourself version.