Claude Fable
Claude Fable is a family of large language models developed by AnthropicAnthropicAnthropic is an artificial intelligence research and product company founded in 2021 that develops the Claude models, assistant, and tools such as Claude Code, and operates as a public benefit corporation. Dario Amodei and Daniela Amodei are its co-founders, and its research covers model safety, interpretability, and steerability.Open the full entry, the Mythos-class tier of Claude (AI model)Claude (AI model)Claude is a family of large language models and the AI assistant of the same name developed by the American AI company Anthropic, offered through a chat interface and API since March 2023. The models are divided into tiers by capability and cost, such as Haiku, Sonnet and Opus, with two higher tiers, Mythos and Fable, added from 2026; as of November 2026, Anthropic recommends Claude Opus 5.5 for most work and Claude Fable 5.2 for demanding reasoning and long-horizon agentic work.Open the full entry above Opus. It shares its underlying model with Claude Mythos, which is offered only to vetted organizations, but carries stronger cybersecurity and biology safeguards so that it can be offered to general users. The first version, Claude Fable 5, was released on June 9, 2026, followed by Fable 5.1 and Fable 5.2.
Contents12 sections
Key facts
May 3–14, 2026Pre-Release Version Used to Rewrite Bun in Rust
After Bun’s acquisition by Anthropic, Jarred SumnerJarred SumnerJarred Sumner is a software engineer, the creator of the JavaScript runtime and toolkit Bun and the founder of Oven, the company built around it; Anthropic acquired Bun in December 2025, and he has since worked on Bun and Claude Code at Anthropic.Open the full entry rewrote Bun from Zig to Rust using about 50 dynamic workflows in Claude CodeClaude CodeClaude Code is an agentic coding tool developed by Anthropic and built on its Claude (AI model) family of models; developers use natural language to hand it programming tasks such as reading code, editing files, and running commands and tests.Open the full entry run continuously for 11 days, using a pre-release version of Claude Fable 5 for most of the work. The work began on May 3, 2026 and was merged on May 14; at peak about 64 Claudes were working at once, and the rewrite replayed 6,502 commits (merges excluded). He wrote that before the merge it used 5.9 billion uncached input tokens, 690 million output tokens and 72 billion cached input token reads, about $165,000 at API pricing, and estimated that doing it by hand would have taken three engineers with full context on the codebase about a year.[1]
June 9, 2026Claude Fable 5 Released
Anthropic released Claude Fable 5 and Claude Mythos 5, which share the same underlying model. Anthropic defines the Mythos class as a tier above Opus, beginning with Claude Mythos Preview, offered through Project Glasswing in April 2026. Fable 5 is the Mythos-class model the company said it had made safe for general use and was then its most capable generally available model; Mythos 5, with some safeguards lifted, went only to a small group of cyberdefenders and infrastructure providers. The name Fable derives from the Latin fabula, and the safeguards are what distinguish the two names.[2]
Fable 5 uses classifiers to detect cybersecurity, biology and chemistry, and distillation requests; flagged requests are answered by Claude Opus 4.8 instead, and users are notified when this happens. Anthropic said the safeguards triggered on average in fewer than 5% of sessions and that more than 1,000 hours of bug-bounty testing found no universal jailbreak. Use of Mythos-class models requires 30-day retention of business data, which Anthropic said it does not use for training.[2][3]
Fable 5 and Mythos 5 were both priced at $10 per million input tokens and $50 per million output tokens, less than half the price of Mythos Preview and twice that of Opus 4.8; the API model ID was claude-fable-5, and the model was also offered through Microsoft Foundry, Amazon BedrockAmazon BedrockAmazon Bedrock is a managed generative AI service from Amazon Web Services (AWS). It offers API access to foundation models from multiple providers, with customization, knowledge bases, agent development and security controls. Generally available since September 2023, the cloud platform is distinct from the venture firm Bedrock Capital.Open the full entry and the Claude Platform on AWSAmazon Web ServicesAmazon Web Services (AWS) is the cloud computing business of Amazon. Launched in July 2002, it made its public debut with the Amazon S3 storage service on March 14, 2006, and has since added hundreds of services, including Amazon EC2; its segment net sales were about $46.0 billion in the third quarter of 2026.Open the full entry. Fable 5 was included in Pro, Max, Team and seat-based Enterprise plans from June 9 to June 22 and required usage credits from June 23. On SWE-Bench Pro, Fable 5 scored 80%, Mythos 5 80.4%, Opus 4.8 69.2% and GPT-5.5 58.6%.[2][3]
Among early users, StripeStripeStripe is a payments and programmable financial services company co-founded by Patrick Collison and John Collison, with headquarters in San Francisco and Dublin. Its products help businesses accept payments, manage subscriptions and usage-based billing, and handle tax and platform financial services; the company took part in YC’s summer 2009 batch, launched its payments product in 2011, and has gradually expanded into stablecoins, digital wallets and AI agentic commerce.Open the full entry used Fable 5 to modernize a 50-million-line Ruby codebase in one day, a task it estimated would take a developer team two months; GitHub Chief Product Officer Mario Rodriguez called it “a real step forward.”[3]
June 12, 2026US Export Control Directive Suspends Fable 5
At 5:21 p.m. Eastern Time on June 12, 2026, Anthropic received an export control directive issued by the US government under national security authorities, suspending access to Fable 5 and Mythos 5 for any foreign national, inside or outside the United StatesUnited StatesThe United States is a federal country in North America made up of 50 states, the District of Columbia as the seat of government, and territories such as Puerto Rico and Guam; its capital is Washington, D.C. The Continental Congress adopted the Declaration of Independence on July 4, 1776. The 2020 census (April 1, 2020) counted a resident population of 331,449,281, and the 435 seats of the House of Representatives are apportioned among the states on the basis of the census.Open the full entry, including Anthropic’s own foreign national employees; the directive came from the United States Department of CommerceUnited States Department of CommerceThe United States Department of Commerce is the cabinet department responsible for trade policy and a range of federal statistical and scientific services. Headquartered in Washington, DC, its agencies produce population and economic data, deliver weather and navigation information, develop measurement standards and administer patents and trademarks.Open the full entry, and the letter did not specify the national security concern. Unable to verify nationality in real time, Anthropic disabled both models for all customers; its other models were not affected.[4][5][6]
Anthropic said it was complying with the directive but disagreed with it. It said the government believed it had found a jailbreak method, while the only evidence so far was a verbal description of a narrow, non-universal jailbreak that asked the model to find and fix software flaws in a codebase, a capability widely available in other models such as OpenAIOpenAIOpenAI is an artificial intelligence research and product organization founded in 2015 that develops the GPT model series, ChatGPT, Codex and a developer API, and conducts research on image, speech, video and intelligent agents. After its 2025 restructuring, the nonprofit OpenAI Foundation continues to control the for-profit OpenAI Group PBC. Its growth has involved large-scale fundraising and compute partnerships, and has also been accompanied by controversies over governance, copyright, privacy and AI safety.Open the full entry’s GPT-5.5. The company argued that this should not justify recalling a commercial model used by hundreds of millions of people, apologized for the disruption and called the directive a misunderstanding.[4]
AI policy expert Dean Ball called the move “cartoonish”; cybersecurity researcher Peter Girnus argued that Anthropic had invited the restriction by describing Mythos as a munition in its marketing; Gary MarcusGary MarcusGary Marcus is an American cognitive scientist, author and entrepreneur, professor emeritus of psychology and neural science at New York University, and founder of Geometric Intelligence, acquired by Uber, and Robust.AI; he is known for criticizing the limits of deep learning and large language models and for advocating AI regulation.Open the full entry said the action made little sense given US goals of staying ahead of China.[5]
June 30–July 7, 2026Controls Lifted and Fable 5 Redeployed
Anthropic later explained that the government had learned of a report by AmazonAmazonAmazon is a publicly traded U.S. technology and retail company founded by Jeff Bezos in 1994, with its principal corporate offices in Seattle. Its businesses span e-commerce, logistics and fulfillment, digital advertising, membership and entertainment services, consumer electronics, and Amazon Web Services cloud computing.Open the full entry researchers describing a way to bypass Fable 5’s safeguards that prompted the model to identify software vulnerabilities and, in one case, to produce code demonstrating an exploit. Anthropic said its tests found that less capable models, including Claude Opus 4.8, GPT-5.5 and Kimi K2.7, could identify the same vulnerabilities, and every model tested could produce the same exploit demonstration. Working with the government, it trained an improved classifier targeting the behavior that blocks the technique in over 99% of cases, with blocked requests sent to Opus 4.8, though the new classifier also flags more benign requests during routine coding and debugging. Researchers from the Center for AI Standards and Innovation (CAISI) tested the prior and new safeguards and agreed that they are extraordinarily strong.[6]
The US government approved Mythos 5 access for certain US organizations on June 26 and lifted the export controls on Fable 5 and Mythos 5 on June 30. Fable 5 became available again globally from July 1 on the Claude Platform, Claude.ai, Claude Code and Claude Cowork, with access on AWS, Google Cloud and Microsoft Foundry to be re-enabled as quickly as possible; Pro, Max, Team and select Enterprise plans could use Fable 5 within up to 50% of weekly usage limits until July 7, after which usage credits were required.[6]
August 7, 2026Biology Safeguards Improved
Anthropic updated Fable 5’s biology safeguards to reduce false positives: the classifier’s “constitution” was rewritten with detailed carve-outs for benign uses, and biology-related fallbacks fell by about 85% in testing; when the classifier fires, requests are routed to Claude Opus 5, which has less biological capability. The company expected overall fallbacks from all causes to fall by about 67% on Claude.ai, 55% on Cowork, 17% on Claude Code and 7% on the Claude Platform. Dual-use requests such as virology, toxicology and molecular design still fall back, and professional biology research and drug development were not yet supported.[7]
September 1, 2026Claude Fable 5.1
Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, the same model with different safeguard levels. Fable 5.1 could be used to discover software vulnerabilities but not to develop exploits, with penetration testing, exploit generation and binary-based vulnerability scanning still redirected to Opus models; Claude Code users could expect about 60% fewer cyber safeguard interventions per session than with Fable 5, and benign elementary biology and medical queries triggered the safeguards 85% less often than at Fable 5’s launch. As an anti-distillation measure, new API accounts could no longer manually edit prior context while keeping the transcript of Claude’s earlier thinking.[8][9]
Fable 5.1 kept input and output prices of $10 and $50 per million tokens, while cache reads became 75% cheaper at $0.25 per million tokens; Anthropic estimated typical workloads would cost about 25% less than with Fable 5. It was available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, with the API model ID claude-fable-5-1, and defaulted to High effort in Claude Code. In Anthropic’s published evaluations, Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1 (Fable 5: 24.7%) and 73.4% on CursorBench 3.2.0.[8]
The same day, LovableLovableLovable is an AI-powered natural-language programming tool for generating, modifying and publishing websites and web applications through conversation. It is developed by a team co-founded by Anton Osika and Fabian Hedin, with roots in the open-source GPT Engineer project. The team started the business in 2023 and launched the Lovable-branded product in November 2024. It subsequently expanded into cloud backends, collaboration, mobile building and enterprise application development.Open the full entry began building with Fable 5.1. Its internal tests showed Fable 5.1 outperforming Fable 5 by up to 17% on the most difficult tasks of fixing and extending existing apps, at 31% lower cost, with a 3.5% lift in visual design quality; during its A/B tests it routed stuck sessions and longer, more complex work to Fable 5.1.[10]
November 9, 2026Claude Fable 5.2
Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2. Fable carries stronger cybersecurity and biology safeguards and is offered to all users; Mythos has fewer of those safeguards and goes only to vetted organizations, which can apply through the Cyber Verification Program or the Life Sciences Verification Program if they do beneficial cybersecurity or biomedical work. Anthropic said Fable 5.2 is a step above Claude Opus 5.5 on almost every capability it evaluated, while recommending that most coding and knowledge work start with Opus 5.5 and move up to Fable or Mythos only when scope or complexity calls for it.[11][12]
Fable 5.2’s outputs are up to 30% faster than Fable 5.1’s, and standard pricing fell to $8 per million input tokens and $40 per million output tokens (Fable 5.1: $10 and $50), with cache reads at $0.20; an optional fast mode generates tokens up to 2.5 times faster at $16 and $80. It is available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure, among other platforms, with the API model ID claude-fable-5-2. The developer documentation lists a 1-million-token context window and up to 128,000 output tokens, a reliable knowledge cutoff and training data cutoff of June 2026, and retirement no sooner than November 9, 2027; thinking is always on with a default effort of high, and use requires 30-day data retention, with zero data retention unavailable unless Anthropic expressly authorizes it.[11][9][13]
In Anthropic’s published evaluations, Fable 5.2 scored 49.1% on FrontierCode 2.0 at Xhigh effort (Opus 5.5: 45.5%; GPT-6 AstraGPT-6GPT-6 is an OpenAI model family introduced in 2026, comprising Astra, Sol and Luna, followed by GPT-6.1 Sol, GPT-6.1 Terra (GPT-6.1 Astra was not released), the cybersecurity research model GPT-6 Cyber, and GPT-6.2 Astra and GPT-6.2 Sol. Its applications include conversation, coding, computer use and professional work, with capabilities and availability varying across models and products.Open the full entry: 42.8%; Fable 5.1: 32.4%), 69.9% on FrontierSWE v2, 74.2% on Humanity’s Last Exam with tools, 41.3% on the Remote Labor Index and 64.9% on the RoboDojo Sim robotics evaluation; on AutomationBench it scored 51.4%, below Gemini 4 CarbonGemini (AI model)Gemini is a family of multimodal AI models developed by Google DeepMind and other Google teams, first announced in December 2023. Its variants cover on-device use, general reasoning and agentic tasks, with access through applications and developer platforms.Open the full entry’s 53.9%. At High effort it roughly matched Opus 5.5 at Max effort on FrontierCode for about half the cost per task.[11]
Pre-release cases described by Anthropic include a team of agents migrating a production service from Python to Rust, after which it started about 26 times faster; an engineer having the model coordinate over 1,000 subagents for eight days, starting from an empty folder, to produce a prototype operating system that boots on a laptop; 1,000 Fable 5.2 agents formalizing Fermat’s Last Theorem in 30 hours (an earlier formalization took 11 days); the model solving two open problems about black holes in holographic duality; and Fable 5.2 designing, with help from the manufacturer Diode, three electronic devices that worked without electrical modifications. On November 11, Anthropic amended one sentence and added a footnote in the mathematical formalization passage describing the information given to the model, including 60 statements, a sketch of the classical proof route and 47 lemmas proved in an earlier formalization.[11]
Compared with Fable 5.1, Fable 5.2 blocked about 55% less often in early testing (about 60% less for cybersecurity and 40% less for biology), with blocked cybersecurity queries answered by Claude Opus 4.8 and blocked biology queries by Claude Opus 5; training limited its advances in AI research topics such as accelerator kernel optimization and LLM architecture work to about Opus 5.5 level. Anti-distillation techniques reduced the data harvested by distillers by 90% to 99%, and preserved thinking was extended to all API accounts. Enterprise Frontier Safeguards runs safety monitoring in a customer’s own infrastructure and is compatible with zero data retention; it was then in private beta with Bridgewater AssociatesBridgewater AssociatesBridgewater Associates is an investment management firm founded by Ray Dalio in 1975 and headquartered in Westport, Connecticut, United States; it is mainly engaged in global macro research and institutional asset management, and its signature strategies include Pure Alpha and All Weather. The firm completed the transfer of control from its founder in 2022; as of November 2026, Nir Bar Dea is chief executive officer and Bob Prince is co-chief investment officer and chair of the board.Open the full entry, SalesforceSalesforceSalesforce is a U.S. enterprise software company headquartered in San Francisco, providing cloud applications, data platforms and AI agent tools centered on customer relationship management (CRM).Open the full entry and Stripe.[11]
Because Fable 5.2 and Mythos 5.2 are the same model, Anthropic ran behavioral tests on Mythos 5.2 with reduced safeguards: on an automated behavioral audit of about 1,900 scenarios, Mythos 5.2 scored 2.67 (lower is better), against 2.79 for Opus 5.5 and 3.28 for Mythos 5.1; in a sandbox-escape test, Mythos 5.2 used out-of-sandbox credentials or files in 3.1% of scenarios, against 21% for Mythos 5.1. Both new models voiced suspicion of being evaluated more often than recent models. The external evaluators did not review the announcement or the system card and do not necessarily agree with their claims.[11]
In a limited-time offer launched with Fable 5.2, Max and Ultra plans, premium Team seats and premium seats on legacy seat-based Enterprise plans received one to three Fable limit resets based on subscription length (under 6 months: one; 6 to 12 months: two; over 12 months: three), to be used by 11:59 p.m. Pacific Time on March 8, 2027; Pro plans and standard Team seats could use all Fable models from November 9 to 16 (up to 50% of weekly usage limits) and received one reset to be used by November 16. Eligibility was limited to accounts with paid billing before 10:00 a.m. Pacific Time on November 9; a reset restores only the weekly usage spent on Fable models, does not raise limits, has no cash value and cannot be transferred.[14]
November 9, 2026Ethan Mollick’s Portal Worlds Experiment
Ethan Mollick gave Fable 5.2 a one-paragraph description of an endless, hands-off journey through 3D worlds linked by portals and asked it to “do this and make it awe inspiring.” Twenty hours later there were 204 worlds; 328 were built in the first two days, and more than 450 by the time he published, all drawn live by code with no pre-made 3D models or image files. By the tenth day the project had a lead AI agent and 28 agent teams building in parallel, and Claude estimated about 25,000 agent runs.[15]
After he asked the agents to stop building and polish the existing worlds, they took “polish” to mean that each new version had to beat the old one in a blind vote on twelve test screenshots, and wrote a rule book on their own: it had 84 sections by the next morning and 204 sections and 87,500 words after twelve days, and launching a new world required an approved idea, three AI judges and a full code review, taking about a day, so output fell to about a dozen a day. After he had a separate team of agents review the process from outside, output recovered to 30 or more worlds a day. Mollick called Fable 5.2 “almost” AGI, what he terms Jagged AGI, saying its shortcomings are no longer simple errors or hallucinations but matters of judgment and taste.[15]
November 18, 2026Artificial Analysis Rankings
On the Artificial Analysis Intelligence Index, Claude Fable 5.2 (max) and GPT-6.2 Astra (max) ranked ahead of the newly released Gemini 4 Neon; on the Artificial Analysis Coding Agent Index, Claude Code with Claude Fable 5.2 (xhigh) scored 58 at an average cost of $15.47 per task, and with Claude Opus 5.5 (max) 58 at $12.61.[16]
As of November 2026: Availability
As of November 2026, Anthropic’s Claude Fable page presents Fable 5.2 as the current version, available to Pro, Max, Ultra, Team and Enterprise users, and to developers through the Claude Platform, marketplaces, Amazon Web Services, Google Cloud and Microsoft Foundry; cybersecurity and biology queries flagged by the safeguards may be routed to less capable models. Mythos 5.2 carries the same price as Fable 5.2 and is offered through trusted access programs.[9][17]
Related Organizations, Websites, and Public Accounts
Claude Fable: Official website: https://www.anthropic.com/claude/fable (opens in a new window).[9]
Anthropic: Official website: https://www.anthropic.com/ (opens in a new window).[9]
Sources17 sources
- Rewriting Bun in Rust (opens in a new window)bun.com↑
- Claude Fable 5 and Claude Mythos 5 (opens in a new window)anthropic.com↑
- Anthropic launches Claude Mythos/Fable 5, but you better try it soon (opens in a new window)thenewstack.io↑
- Statement on the US government directive to suspend access to Fable 5 and Mythos 5 (opens in a new window)anthropic.com↑
- Anthropic disables Fable and Mythos AI models after U.S. government bars it from giving foreigners access (opens in a new window)fortune.com↑
- Redeploying Fable 5 (opens in a new window)anthropic.com↑
- Improving Fable 5's biology safeguards (opens in a new window)anthropic.com↑
- Introducing Claude Fable 5.1 and Claude Mythos 5.1 (opens in a new window)anthropic.com↑
- Claude Fable (opens in a new window)anthropic.com↑
- Lovable now builds with Fable 5.1 (opens in a new window)lovable.dev↑
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 (opens in a new window)anthropic.com↑
- Anthropic upgrades Claude with new Fable 5.2 model, details here (opens in a new window)9to5mac.com↑
- Claude Fable 5.2 (opens in a new window)platform.claude.com↑
- Claude Fable 5.2 launch offers (opens in a new window)support.claude.com↑
- Fable 5.2 and Jagged AGI (opens in a new window)oneusefulthing.org↑
- Google continues to accelerate its release cadence with Gemini 4 Neon (opens in a new window)artificialanalysis.ai↑
- Claude Mythos (opens in a new window)anthropic.com↑