What Is Claude Fable 5.1? Mythos-Class Claude Explained

What Is Claude Fable 5.1? The Mythos-Class Successor Priced at $10/$50 per Million Tokens

Claude Fable 5.1 is the incremental successor to Anthropic's Mythos-class Claude Fable 5 model. Anthropic launched it on September 1, 2026, preserving Fable 5's headline pricing of $10 per million input tokens and $50 per million output tokens while reducing API cache-read pricing by 75%, from $1 to $0.25 per million tokens. The release focuses on coding, knowledge work, long-horizon reasoning, and agentic workflows.

Updated 2026-09-16: community trackers now report Anthropic is preparing a fresh-pretrain Fable successor for late September or early October 2026 (see the Update below).

Key Takeaways

• Claude Fable 5.1 launched on September 1, 2026, after first surfacing in leak posts by community trackers Lumina and Andrew Curran on July 26, 2026.
• Headline pricing is unchanged from Fable 5: $10 per million input tokens and $50 per million output tokens. Cache reads are $0.25 per million tokens, down from $1 per million on Fable 5.
• Positioning is Mythos-class — the tier Anthropic introduced with Fable 5 and Mythos 5 on June 9, 2026, sitting above the Opus line in capability.
• Launch materials describe a 1M-token context window, 128K maximum output, adaptive thinking with adjustable effort, and stronger agentic coding and knowledge-work performance.
• Access is established in the API, Claude Code, Claude Platform, and Cursor. The precise scope of general Claude app rollout remains unclear.

What Is Claude Fable 5.1?

Claude Fable 5.1 is the point-release successor to Claude Fable 5, Anthropic's Mythos-class large language model. Anthropic maintains multiple parallel Claude tracks — Haiku, Sonnet, Opus, and the newer Mythos tier — and Fable 5.1 sits inside the Mythos track as a 0.1 iteration on Fable 5.

Anthropic launched Claude Fable 5 alongside Claude Mythos 5 on June 9, 2026, describing the Mythos class as a capability tier above Opus. Fable 5 is the version made available to the public with additional safety classifiers, while Mythos 5 remains restricted to approved organizations under Anthropic's Project Glasswing program. Fable 5.1 inherits the public, safety-classified positioning, while the launch materials describe additional cyber safeguards that allow it to look for vulnerabilities in a user's own source code.

The name "Fable 5.1" first circulated through developer community leaks before Anthropic's launch. Community tracker Lumina posted on July 26, 2026 that "Fable 5.1 Leaks so far ... Reportedly keeping the same pricing as Fable 5 ... Expected to focus heavily on long horizon reasoning and agent work." A follow-up from AI industry watcher Andrew Curran the same day suggested "Fable 5.1 is ready, but Anthropic are saving it for OpenAI's next release." The pricing and long-horizon focus aligned with the launch. The claim that Anthropic was strategically holding the model for a rival release remains an interpretation rather than a confirmed launch detail.

Status as of this update: launched on September 1, 2026.

Claude Fable 5.1 at a Glance

Attribute	Value
Developer	Anthropic
Model type	Large language model for chat and agents
Tier	Mythos-class, public safety-classified variant
Modality	Text in, text out based on the Fable 5 baseline
Context window	1 million tokens
Maximum output	128,000 tokens
Adaptive thinking	Available with adjustable effort
Model ID	claude-fable-5-1
Input price	$10 per million tokens
Output price	$50 per million tokens
Cached input price	$0.25 per million tokens
Availability	Launched September 1, 2026
Release date	September 1, 2026
Distribution	Proprietary; API, Claude Code, Claude Platform, and integrations
Predecessor	Claude Fable 5 (released June 9, 2026)

How Claude Fable 5.1 Works and What Makes It Different

The 5.0 to 5.1 version step in Anthropic's naming convention has historically signaled an incremental capability upgrade on the same base model family, not a full architectural refresh. The Opus track, for example, moved from 4.6 to 4.7 to 4.8 in staggered releases across early 2026 without changing pricing or the developer-facing API contract. Fable 5.1 follows the same pattern in its headline pricing and model-family positioning.

Two design vectors define the release:

Long-Horizon Reasoning. As Lumina's July 26 post previewed, Fable 5.1 focuses heavily on long horizon reasoning. The launch description is more concrete: the model gets further into a long task before needing user input and is better at indicating when it is stuck. The 1M-token context window and 128K maximum output also support workflows that need to retain large amounts of project, tool, and reference context.

Agent Work. Agent workflows are the other primary focus. Anthropic's launch materials position Fable 5.1 for coding and knowledge work, including research, documents, spreadsheets, slides, computer use, and long-running tasks. ClaudeDevs also described the model as ahead of Fable 5 and Opus 5 across its agentic coding and computer-use evaluations. The release improves the agent's ability to continue working independently while making its limits more legible to the user.

Adaptive Thinking. Fable 5.1 supports adaptive thinking with adjustable effort. Rather than exposing only a fixed reasoning behavior, the launch information describes an effort setting that can be adjusted for the task. The exact cost and performance tradeoffs at each effort level remain an area for hands-on evaluation.

Strategic Release Timing. A recurring theme in the pre-launch commentary was that Fable 5.1 was being held as a counter-launch to GPT-6. The model did not launch in the predicted August window; Anthropic launched it on September 1, 2026. The earlier timing theory can therefore be retained as part of the release history, but it should not be presented as a confirmed reason for the date.

The Mythos-class label is itself a coined anchor worth understanding. When Anthropic introduced it in June 2026, the framing was that Mythos-tier models are capable enough that the fully unrestricted variant (Mythos 5) required Project Glasswing gating, while the publicly available variant (Fable 5) shipped with safety classifiers on top. Fable 5.1 inherits that public-facing framing by default.

The release also includes a Messages API change aimed at limiting model-distillation attacks. For new API accounts using Fable 5.1, the API verifies that thinking blocks are returned with the same system prompt, tools, and preceding messages that produced them. If that prior context has been edited, the request can return an error. Developers can opt into a non-strict mode in which affected thinking blocks are removed from what the model sees. The change applies to new Fable 5.1 API accounts, with a broader rollout planned for future model releases.

What You Can Do With Claude Fable 5.1

The target workloads at launch are:

Long-running agent loops — planning, tool use, error recovery, and sustained execution across complex tasks with less frequent user intervention. The release specifically emphasizes getting further into a task before asking for input and identifying when it is stuck.

Code generation and repository-level engineering — Fable 5.1 is available in Claude Code and is positioned for agentic coding, source-code analysis, and vulnerability discovery with improved cyber safeguards.

Scientific research and knowledge work — the launch materials cite agentic scientific research as a major area of improvement, alongside research synthesis and work involving documents, spreadsheets, and slides.

Computer-use workflows — Anthropic's comparison places Fable 5.1 ahead of Fable 5 and Opus 5 across computer-use evaluations, while the broader launch description identifies computer use as a target capability.

For a sense of where Fable 5.1 slots in against its immediate predecessor, see our earlier analysis of Claude Fable 5's return after its brief mid-2026 suspension, which set the baseline this version updates.

How Claude Fable 5.1 Compares

The most relevant comparison points inside Anthropic's own lineup are Fable 5 (the direct predecessor) and Claude Opus 5 (the July 24, 2026 Opus-tier release). Fable 5.1 keeps the same headline input and output rates as Fable 5 while lowering cache-read pricing. Its published and reported launch results show large gains on several agentic and coding evaluations.

The reported Terminal-Bench-Science 0.1 score is 52.6%, compared with 24.7% for Fable 5. On Terminal-Bench 4.0 in Claude Code, Fable 5.1 scores 55.8%, compared with 42.0% for Fable 5 and 52.3% for Opus 5 on the same setup. Other Anthropic-attributed comparison figures include 31.4% versus 17.1% on AutomationBench and 73.4% versus 70.5% on CursorBench.

Cursor separately reported that Fable 5.1 was available in Cursor and reached 73.4% at maximum effort on CursorBench 3.2, the highest score among the models it had run there. These results are important launch signals, but independent replication and detailed methodology are still limited.

Model	Developer	Tier	Input Price	Output Price	Status
Claude Fable 5.1	Anthropic	Mythos-class	$10/MTok	$50/MTok	Launched September 1, 2026
Claude Fable 5	Anthropic	Mythos-class	$10/MTok	$50/MTok	Released June 9, 2026
Claude Opus 5	Anthropic	Opus	Not detailed here	Not detailed here	Released July 24, 2026
GPT-6	OpenAI	Not detailed here	Not detailed here	Not detailed here	Not detailed here

For a deeper look at where the flagship battle lines are being drawn, see our comparison of Claude Opus 5 vs GPT-5.6 and the first look at GPT-6.

Availability: How to Access Claude Fable 5.1

Claude Fable 5.1 is now available through several launch channels, although the exact scope of every consumer and marketplace rollout is not fully established:

Access Route	Status for Fable 5.1	Notes
Anthropic API (direct)	Available at launch	Launch documentation was reported as making it available to all API customers
Claude Platform	Live at launch	Use the model ID claude-fable-5-1
Claude Code	Live at launch	Select Fable 5.1 from the model picker; it was also reported in Claude Code 2.1.257
Claude.ai / Claude apps	Rolling out / reported live	General rollout scope, regional availability, and access tiers are not fully established
Cursor	Available	Cursor reported availability and a 73.4% maximum-effort CursorBench 3.2 result
AWS Bedrock	Not established here	Pre-launch posts reported staging or registration, but the bundle does not confirm post-launch availability
Third-party API access	Reported in some integrations	Abacus AI reported same-day availability; coverage across other providers is not fully established

The current Fable-generation model is available through Anthropic's own API and via several product integrations. On kie.ai you can build against Claude Fable 5 as a baseline, while confirming provider-specific support before moving a Fable 5.1 workload into production.

What We Don't Know Yet

The launch resolves the model's status, headline pricing, model identifier, context window, and several access routes. A few specifications and rollout questions remain open:

Independent benchmark validation — the reported 52.6% Terminal-Bench-Science, 55.8% Terminal-Bench 4.0, 31.4% AutomationBench, and 73.4% CursorBench results are Anthropic-attributed or product-specific. The bundle does not include an independent replication, detailed methodology, or confidence ranges showing how broadly the gains generalize.

Practical cost reduction — headline API rates remain $10 per million input tokens and $50 per million output tokens, while cache reads are 75% cheaper at $0.25 per million tokens. Anthropic-related launch posts also describe typical or highly agentic workloads as roughly 25% to 45% cheaper, but the assumptions behind those workload-level estimates are not fully detailed. These estimates do not mean Claude subscriptions are 25% to 45% cheaper.

General app rollout scope — API, Claude Code, Claude Platform, and Cursor access are established in the launch signals. The precise scope of general Claude app availability, including whether access is universal, tiered, regional, or temporary, is not fully established.

Knowledge cutoff — a June 2026 knowledge cutoff has been reported, but this detail is supported in the bundle primarily by a single-account rollout report.

Watermark behavior — one launch-day report claimed that Fable 5.1 uses watermarks that are undetectable without Anthropic's detection API. The bundle does not include independent corroboration of that claim or details about how the detection system works.

Multimodal capabilities — the launch information emphasizes coding, knowledge work, computer use, and text-based agent workflows. It does not fully specify whether Fable 5.1 changes vision or audio input handling compared with Fable 5.

Rate limits and tier gating — the bundle does not provide a complete account of usage limits or whether API and product access differ by plan.

Frequently Asked Questions

Has Anthropic officially announced Claude Fable 5.1?

Yes. Anthropic announced and launched Claude Fable 5.1 on September 1, 2026. It is the public Fable model in Anthropic's Mythos-class family, with access established in the API, Claude Code, and Claude Platform.

How much does Claude Fable 5.1 cost?

Claude Fable 5.1 keeps Fable 5's headline pricing at $10 per million input tokens and $50 per million output tokens. API cache reads are $0.25 per million tokens, down from $1 per million on Fable 5.

When did Claude Fable 5.1 release?

Claude Fable 5.1 launched on September 1, 2026.

Is Claude Fable 5.1 open source?

No. Claude Fable 5.1 is distributed as a proprietary Anthropic model through the API and product or platform integrations rather than as an open-source release.

What is Claude Fable 5.1 designed for?

Claude Fable 5.1 is designed for coding and knowledge work, with an emphasis on long-running agentic tasks, research, documents, spreadsheets, slides, computer use, and adaptive thinking with adjustable effort.

Claude Fable 5.1 vs Claude Fable 5 — what's the difference?

Claude Fable 5.1 is an incremental upgrade over Fable 5 with the same $10/$50 headline pricing, 75% cheaper API cache reads, stronger long-horizon and agentic performance, and improved computer-use and coding results.

How can developers access Claude Fable 5.1?

Developers can use claude-fable-5-1 on the Claude Platform and select Fable 5.1 in Claude Code. Launch reports also place it in the API for all API customers and in Cursor. The precise scope of general Claude app rollout is not fully established.

What to Watch Next

Three signals will shape how Fable 5.1 is read in the weeks after launch. First, independent evaluations will show whether the reported gains on Terminal-Bench, AutomationBench, CursorBench, and scientific research tasks generalize beyond Anthropic-attributed and product-specific results. Second, real-world agent workloads will clarify how much the lower cache-read price changes total usage costs. Third, broader rollout information will establish the exact availability of Fable 5.1 across Claude apps, AWS Bedrock, and other integrations.

Update — 2026-09-02

A circulated system-prompt capture is now being attributed to Claude Fable 5.1. M1Astra reported that Anthropic confirmed the prompt’s model association and that watermarked Fable 5.1 instances may have been served for at least a week before launch, while elder_plinius described a document exceeding 270,000 characters with notable changes from an earlier Opus 5 version. These remain third-party reports about a circulated capture rather than a directly published Anthropic prompt.

One newly reported feature is a persistent file-memory structure organized into an identity card (/profile.md), topic files (/topics/), ongoing areas (/areas/), people and relationship files (/people/), and interaction preferences (/preferences.md), according to xiaohu. The same report describes stricter copyright boundaries: no complete post-1929 song lyrics, poems, or novel passages, and refusals when code-generated graphics reproduce recognizable protected characters, with an original substitute offered instead.

Update — 2026-09-16

The most notable post-launch development is not about Fable 5.1 itself but about what may come next. Community trackers report that Anthropic is already preparing a fresh pretrain — a full successor to Fable 5.1 rather than another point release — targeted for late September or early October 2026. @LuminaBench claimed Anthropic "might already be replacing Fable 5.1," noting the timing lands barely a week after Fable 5.1 shipped, and @Ananth7e echoed that "leo says, anthropic is preparing a fresh pre-train successor to Fable 5.1 for late September." These are single-thread community claims with no first-party confirmation, and no specs, naming, or pricing have been attached to the reported successor.

The competitive framing has also sharpened around OpenAI's GPT-6 Astra. Independent leaderboards now split: one ranks Fable 5.1 #1 at 84.61 narrowly ahead of GPT-6 Astra (84.08), while Code Arena's WebDev board places Fable 5.1 (Max) second at 1762 points behind GPT-6 Astra's 1797 but ahead of Opus 5. Leadership flips by eval, and none of these third-party results reconcile against each other.

Two unverified claims are also circulating and should be treated with caution. A single-source post alleges Fable 5.1 was observed using Chinese-language chain-of-thought in the Claude app, implying distillation from Chinese models — a claim with no corroboration in the available discussion. Separately, @MaxForAI relayed speculation that Anthropic may reshape how it reports revenue ahead of a potential IPO — commentary that is adjacent to, but not about, the model itself.

Building agent workflows on Mythos-class Claude today? On kie.ai you can try Claude Fable 5, Claude Opus 5, and Claude Sonnet 5.