Claude Fable 5.1: pricing and how it differs from Opus 5.5

Claude Fable 5.1 remains the top-tier model in Anthropic's public lineup as of 29 October 2026: a 1M-token context, up to 128K tokens in a reply, knowledge through June 2026. You pick it for hard reasoning and long agentic work. For everything else, Anthropic itself advises starting with Opus 5.5, which is cheaper and faster.

What Claude Fable is and where it came from

Fable belongs to the Mythos class, which sits above Opus in Anthropic's lineup. The name "Fable" comes from the Latin fabula, and "Mythos" from a Greek word with the same meaning. As Anthropic explains it, the names reflect a difference in safeguards, while the capabilities of the models are the same. Fable is available to everyone, with restrictions in high-risk areas (cybersecurity, biology). Mythos, with relaxed restrictions, goes to vetted organizations through special programs, and access to Mythos 5.1 is so far open only to organizations in the US.

The model has a short and bumpy history:

• 9 June 2026: Fable 5 and Mythos 5 are released.
• 12 June: because of US export restrictions, Anthropic suspends both models for all users.
• 1 July: Fable 5 is available worldwide again.
• 1 September: Fable 5.1 is released, and Fable 5 gets legacy status.

All the dates come from Anthropic's note on redeploying Fable 5 and the Fable 5.1 page in the documentation.

Fable 5 and Fable 5.1: what changed

The base price did not change, but cache reads got four times cheaper. That matters for agents: they loop over the same context again and again, and the cache makes up a noticeable share of the bill. By Anthropic's estimate, Fable 5.1 costs about 25% less than Fable 5 on typical workloads and up to 45% less on heavily agentic ones. That is the vendor's claim, and we did not verify it independently.

Three more changes from the announcement. Safeguards trigger less often: in Claude Code sessions, cyber restrictions step in about 60% less. The model may be used to find vulnerabilities in source code, but not to develop exploits, and some dual-use tasks such as pentesting still go to Opus models. For API developers there are breaking changes: a forced tool call returns an error, and reasoning blocks are tied to the model that produced them. The full list is in the migration guide on the model's page.

Fable, Opus 5.5 and Sonnet 5.5 compared

Next, Anthropic's own numbers. In the table from the Opus 5.5 announcement, the new model beats Fable 5.1 on all nine benchmarks shown. Some of the rows:

The figures come from the Opus 5.5 announcement. Opus 5.5's results were obtained with adaptive thinking at maximum effort, while the table does not state the conditions for Fable. Anthropic itself notes that the benchmark gap has become a less reliable guide to real work. The takeaway: on standard tests Fable 5.1 does not beat the new Opus, and the documentation leaves Fable the role of the model for hard reasoning and long agentic work. Test on your own repository.

The three tiers are compared in detail in Opus 5.5, Sonnet and Fable: which to choose. On Opus alone: Claude Opus 5.5: free or paid, API price and access.

What Fable 5.1 costs at Anthropic and at codexkey

codexkey prices as of 29 October 2026, premium channel. Current numbers are on the models page. Two points are worth noticing. In the codexkey catalog Fable 5.1 costs about 8.7 times more than Opus 5.5 on both input and output, while at Anthropic the gap is 2.5 times. And the cache read for Fable 5.1 in the codexkey price list (0.76) is higher than Anthropic's official 0.25. For long agentic sessions you need to account for that in your estimate.

When Fable pays off

Count the price per solved task, not per request: divide the cost of an attempt by the share of tasks the model solves. Say one agentic attempt spends 1M input and 50K output tokens with no cache. At Anthropic that is $5.00 on Opus 5.5 and $12.50 on Fable 5.1, and in the codexkey price list $1.08 and $9.35. The assumptions are illustrative: this is arithmetic, not a measurement.

The formula gives a threshold: Fable pays off on tokens only if it solves as many times more tasks than Opus as it costs more. With a 2.5x price gap, Opus at a 30% success rate requires at least 75% from Fable. With an 8.7x gap (as in our catalog), Opus at 10% requires 87% from Fable. If Fable solves fewer, Opus is cheaper per solved task.

So Fable is bought for another reason. It is for cases where a mistake costs more: an engineer's hour, a broken release, the collapse of a long multi-hour chain. Anthropic itself states the threshold this way: move to Fable when your tests on Opus 5.5 with raised effort still do not give the result you need. We covered how to build such a test set in how to compare models on your own tasks.

Where Fable 5.1 is available and how to turn it on

According to Anthropic's help page on Fable on your plan, the model is on all paid plans, with different terms:

• Free: Fable is unavailable, and there is no free access to it in chat.
• Pro, Standard seats in Team and Enterprise: Fable runs on usage credits, meaning pay-as-you-go on top of the subscription.
• Max, Premium seats in Team and Enterprise: you can spend up to 50% of the weekly limit on Fable without extra payment, and credits apply after that.
• API and usage-based Enterprise: standard per-token rates.

In Claude Code, Fable is not the default on any plan. The default is now Opus 5.5, and you turn Fable on with the /model fable command or the claude --model fable flag (version 2.1.257 or newer is required). Before it spends credits, the program shows a confirmation prompt. Details are in the Claude Code model documentation.

The query "claude fable 5.1 max" is ambiguous. It can mean the Max plan (where Fable counts toward 50% of the weekly limit) or the max effort level, which Fable 5.1 supports alongside low, medium, high and xhigh.

Fable 5.1 from Russia through codexkey

Anthropic plans cannot be bought from Russia: the country is not on the supported list, and the details are in Claude API from Russia without a VPN. Through the codexkey API gateway the model is available as claude-fable-5-1: a sk_cdx_... key, payment in rubles by a Russian card, through SBP (the Russian Faster Payments System) or with crypto, with no VPN and no Anthropic account. We do not sell a Claude Max or Pro subscription; Anthropic's plan prices are collected in how much a Claude subscription costs.

A call through the Anthropic SDK. The address is given without /v1, and set max_tokens with a margin, because thinking is always on in Fable:

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.codexkey.online",
    api_key="sk_cdx_YOUR_KEY",
)

message = client.messages.create(
    model="claude-fable-5-1",
    max_tokens=4096,
    messages=[{"role": "user", "content": "Find the data race in this code and suggest a patch: ..."}],
)
print("".join(block.text for block in message.content if block.type == "text"))

In Claude Code the model is turned on with /model claude-fable-5-1[1m], and the [1m] suffix opens a 1M-token window. Gateway setup is described in the Claude Code guide, and beginners will find Claude Code: install, pay and pricing useful.

Rumors about Fable 6 and Fable 5.5

People search for "claude fable 6" and "claude fable 5.5". In Anthropic's model overview as of 29 October 2026, the current models are Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, and the deprecated list holds Fable 5. There is no Fable 5.5 or Fable 6 there, and we found no announcements. The number "5.5" is so far worn by the other three lines: Opus 5.5 came out on 22 September, Haiku 5.5 appeared on 7 October. Treat anything else as rumor until Anthropic publishes a model page.

FAQ

What is Claude Fable in simple terms?

It is Anthropic's strongest public model, ranked above Opus in the lineup. It was built for long agentic tasks and hard reasoning, and the price for that is $10 and $50 per million tokens and higher latency.

How does Fable 5.1 differ from Fable 5?

The input and output prices are the same, but cache reads dropped from $1 to $0.25 per million tokens. Knowledge is updated through June 2026 instead of January, safeguards trigger less often, and the API got breaking changes, for example an error on a forced tool call.

Is Fable better than Opus 5.5?

Not always. In Anthropic's own table Opus 5.5 beats Fable 5.1 on the nine benchmarks shown, and it costs 2.5 times less. Trying Fable makes sense where Opus with raised effort fails on your tasks.

Is there a free Claude Fable?

No. The Free plan does not include Fable, Pro gets the model through usage credits, and Max includes up to 50% of the weekly limit. The API charges per token. We found no free way to get Fable without breaking Anthropic's rules, so steer clear of dubious "free keys".

Do Fable 6 and Fable 5.5 exist?

Anthropic's documentation as of 29 October 2026 has no such models, so these are rumors. Keep an eye on the models page at platform.claude.com.

How do I buy Claude Fable from Russia?

An Anthropic subscription cannot be set up from Russia, because the country is not on the supported list. The working route for developers is the API through a gateway with payment in rubles. At codexkey the model is called claude-fable-5-1; check the price on the models page.