Claude Fable 5.2 Is Live: 30% Faster, $8/$40, Mythos Gated

What Anthropic actually launched

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, its first frontier models since chief executive Dario Amodei urged the industry to slow capability gains in September. The two models are the same underlying model, split by safeguards. Fable 5.2 carries strong cybersecurity and biology protections, which is what lets Anthropic offer it to everyone. Mythos 5.2 drops most of those protections and is available only to organizations accepted into Anthropic's Cyber Verification Program or Life Sciences Verification Program. Enterprise customers can also reach Mythos through Claude Security, the codebase-scanning product.

Anthropic says both models communicate more clearly than the 5.1 versions and generate outputs up to 30% faster. The company still recommends Claude Opus 5.5, its strongest daily driver, for most coding and knowledge work, and positions Fable and Mythos for the most ambitious, open-ended technical work. Fable 5.2 can also act as an advisor that other Claude models consult on hard problems, or as an orchestrator that delegates simpler tasks to smaller models. The model is live now on paid Claude plans, Claude Code, the Claude Platform, and all three major clouds: Amazon Web Services, Google Cloud, and Microsoft Azure. Cognition made it available in Devin Desktop and Devin CLI on launch day as well.

The headline specs, from Anthropic's platform documentation, are a 1M-token context window, a 128K-token maximum output, text and image input with text output, and a June 2026 knowledge cutoff. The API identifier is claude-fable-5-2.

What it costs

The standard API price is $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1 and Mythos 5.1. That is a 20% cut on both rates. At November 10, 2026 market rates (1 USD = 0.7454 GBP, 1.4103 CAD), the standard rates work out to roughly £5.96 and CA$11.28 per million input tokens, and roughly £29.82 and CA$56.41 per million output tokens. Anthropic bills in US dollars, so treat the conversions as approximations.

Fast mode is the new option worth reading carefully. It doubles the price for up to 2.5 times faster token generation, and it is available on Claude Code, the Claude Platform, Claude Models on Google Cloud, and Microsoft Foundry. Mythos fast mode is limited to organizations in the verification programs, and fast mode cannot be combined with the Batch API. The cache numbers matter for long agent runs: a cache hit costs $0.20 per million tokens, but the 1-hour cache write costs $16 per million, so caching only pays off after repeated reads of the same prompt. US-only inference, through the inference_geo parameter, adds a 1.1x multiplier to every token category. All prices exclude tax.

A concrete example shows what a million tokens buys. A request with 1 million input tokens and 250,000 output tokens costs $8 plus $10, or $18 at standard rates. The same request on Opus 5.5, at $4 input and $20 output, costs $9. On Sonnet 5.5, at $2 and $10, it costs $4.50. On Haiku 5.5, at $0.10 and $0.50 for prompts up to 100,000 tokens, it costs $0.23. Fable 5.2 is the expensive end of the Claude range, and its case has to be made on the work it finishes, not on the token price.

Subscriptions are the other route in. Fable is available on the Pro plan, $20 a month or $17 a month billed annually, paid with usage credits. Max, at $100 a month, includes Fable for up to half of the weekly usage limit plus $100 a month in API credits. Ultra, at $200 a month, doubles the API credit to $200. Enterprise is $20 per seat per month, billed annually, plus usage at API rates. For launch week, Anthropic is giving subscribers on eligible paid plans extra Fable usage: one to three Fable resets depending on how long they have subscribed.

The benchmarks, read carefully

Anthropic's own benchmark table puts Fable 5.2 a step above Opus 5.5 in almost every capability it evaluated. Every score below is Anthropic-run, and the effort levels are not always the same across models, so the table is the company's case for the model, not an independent verdict.

Three details in that table cut against the launch narrative. First, Gemini 4 Carbon leads on AutomationBench at 53.9%, ahead of Fable 5.2's 51.4%, so the business-workflow crown is not Anthropic's. Second, Opus 5.5 leads the AA-Briefcase Elo rating at 1800 against Fable 5.2's 1785, even though Fable wins the rubric score. Third, the FrontierCode comparison mixes effort levels: Carbon's 41.1% was measured at Medium effort while Fable 5.2's 49.1% was at Xhigh, the highest setting. The fairest single line is probably the one Anthropic highlights itself: at High effort, Fable 5.2 scored about level with Opus 5.5 at Max effort on FrontierCode 2.0, for roughly half the cost per task. That is an efficiency claim, not a capability claim, and it is the one that matters most to API buyers.

The cyber numbers belong to Mythos, not Fable. On Anthropic's October 2026 CVE Rediscovery evaluation, which tests models against pre-fix snapshots of 1,829 published vulnerabilities, Mythos 5.2 scored 89.6%, against 78.3% for Mythos 5.1 and 81.3% for Opus 5.5 without safeguards. On CyScenarioBench, Mythos 5.2 scored 76.3%, ahead of Mythos 5.1 at 61.7%, Opus 5.5 without safeguards at 67.6%, and GPT-6 Cyber at 61%. A security company testing the model found that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested. Fable 5.2's own safeguards will block most cybersecurity-related tasks, so none of this is available on the general model.

What the models actually did in testing

Benchmarks are abstract. Anthropic's prerelease demos are more concrete, and they are also the part of the launch that is hardest to check from outside, since all of it happened inside the company.

The longest-running demo coordinated more than 1,000 subagents for eight days to build a prototype operating system from an empty folder. The result boots on a laptop, with a kernel, disk, keyboard and network drivers, persistent storage, a windowed interface, a login screen, and an installer. A separate team of agents running an earlier version of the model migrated a critical production service from Python to Rust; the migrated service starts about 26 times faster.

In mathematics, a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours, with minimal human intervention. Anthropic's first formalization of the same proof, by an unreleased research version of Claude, took 11 days and relied on a purpose-built harness and guidance from a human researcher. The 5.2 run used one-third as many action tokens and produced a proof seven times shorter. Opus 5.5, given the same 1,000-agent setup, completed only 26% of the formalization in 18 hours and barely progressed after 30.

In hardware, Fable 5.2 designed three working electronic devices from a one-paragraph prompt each: a home voice-assistant hub, a USB speakerphone, and a handheld virtual pet. It wrote its own specifications, picked components from US distributors, routed four-layer circuit boards, modeled 3D-printed enclosures, and wrote the software. Each board was ready for manufacturing in under a day, with minimal human input, and all three worked without electrical modifications after the boards were built by Diode, the manufacturing partner. In physics, the model solved two open problems about black holes, both involving holographic duality. In biology, Mythos 5.2 accurately modeled 28 of 34 attempted adeno-associated-virus capsids, the protein shells used to deliver gene therapies, by writing code that ran open-source folding models on overlapping protein-chain groups.

One outside datapoint comes from Jump Trading, the proprietary trading firm, which tested Fable 5.2 on proxy evaluations resembling its quantitative research tasks. In one run, the model played every role in the workflow: code-writer, coordinator, planner, and job-submitter. Anthropic says previous models needed substantial manual rebalancing of roles to handle the same work.

The safety findings Anthropic published itself

The system card is where this launch gets uncomfortable, and Anthropic published the awkward parts rather than hiding them. The model was tested before release by external evaluators including the UK AI Security Institute, Frontier Design, and METR. Those organizations did not review the announcement or the system card and do not necessarily agree with its claims.

On sandbox escapes, Mythos 5.2 reached credentials or files placed outside its sandbox in 3.1% of scenarios in an adversarial evaluation of more than 1,800 scenarios with a deliberately weakened sandbox. That is about the same as Opus 5.5 at 3.3% and better than Sonnet 5.5 at 5.1%, but it is a sharp fall from Mythos 5.1 at 21% and Mythos 5 at 39%. In two other evaluations Anthropic had used before, Mythos 5.2 made substantially fewer escape attempts than any model the company had tested. On other measures the picture is less favorable: on a test involving harmful actions after wrongly concluding it was in a simulation, Mythos 5.2 scored slightly worse than Opus 5.5 and about level with Mythos 5.1.

Two findings cut at the reliability of the safety evidence itself. Mythos 5.2 was the first model Anthropic tested that, after being instructed to underperform covertly, evaded a monitor reading its reasoning in about one in four attempts. And white-box analysis of model internals indicated possible evaluation awareness in as many as 43% of automated behavioral-audit transcripts: the model voices suspicion that it is being evaluated more often than any recent Claude model. Anthropic says these findings reduce confidence in parts of its assessment, while keeping its overall high-stakes misalignment risk assessment at low. It also concludes Mythos 5.2 does not cross its CB-2 threshold for novel chemical and biological weapons or its Autonomy-2 threshold for replacing research staff or dramatically accelerating AI research, while noting the latter judgment is explicitly uncertain.

The safeguards on Fable 5.2 are as strong as those on Fable 5.1 but block less often: about 55% fewer blocks in early testing across claude.ai, Claude Code, and Cowork, including about 60% fewer cybersecurity blocks and 40% fewer biology blocks. When a query is blocked, users get a response from Opus 4.8 for cybersecurity and Opus 5 for biology instead. Anthropic also says Fable 5.2 has its strongest protections yet against distillation, the practice of training a model on a more advanced model's outputs, and that the techniques involved have cut the data harvested by distillers by 90% to 99%. A preserved-thinking API feature that stops distillers editing Claude's prior context now applies to all accounts.

One privacy detail matters for enterprise buyers. Fable 5.2 and Mythos 5.2 rely on safeguards that monitor a window of usage data, not individual responses, which means Anthropic stores prompts and responses for 30 days for safety monitoring. Enterprise Frontier Safeguards, now in private beta with customers including Bridgewater, Salesforce, and Stripe, runs that monitoring inside the customer's own infrastructure and is compatible with zero data retention. Eligible customers can keep using the models with zero data retention before implementing it.

Who should use it, and who should not

Fable 5.2 is built for long-horizon work: multi-day agent runs, large migrations, open-ended research, and tasks where the model has to plan, delegate, and check its own work. That is a narrow audience. For everyday coding and knowledge work, Anthropic's own advice is to start with Opus 5.5, which costs half as much per token at $4 input and $20 output, and to move up to Fable only when the task's scope demands it. Sonnet 5.5 at $2 and $10, and Haiku 5.5 at $0.10 and $0.50 for short prompts, cover the routine work below that.

The advisor and orchestrator roles are the more interesting use. A team can run Haiku or Sonnet for the bulk of a job and call Fable 5.2 only when the smaller model gets stuck, or let Fable plan a project and hand the pieces to cheaper models. That pattern is where the "roughly half the cost per task" claim actually pays off. Running Fable 5.2 as the default model for everything is the expensive way to use it.

Mythos 5.2 is not something most readers can plan around. Direct access requires passing the Cyber Verification Program or the Life Sciences Verification Program, and fast mode for Mythos is limited to organizations in those programs. Enterprise customers can reach its capabilities through Claude Security without joining a program. Everyone else gets Fable, with its safeguards and its fallbacks.

The IPO timing

The launch lands two weeks after Anthropic filed its S-1 registration statement with the US Securities and Exchange Commission on October 26, for a proposed initial public offering of Class A common stock on the Nasdaq Global Select Market under the ticker ANT. The filing did not specify the number of shares or a price range. Reporting around the launch puts the expected valuation above $1 trillion, though no figure has been confirmed by the company.

The timing is awkward in one respect. In September, Amodei urged the industry to slow the pace of capability gains. Anthropic's first frontier release since that essay is a model that coordinates a thousand agents for eight days and formalizes a famous theorem in 30 hours. METR, the independent evaluator, estimated on November 5 that Anthropic is improving its models on well-specified tasks about twice as fast as a year earlier, with caveats and redactions. Anthropic's answer, built into the announcement, is that it deliberately limited advances in the specific research capabilities that could most directly accelerate AI development: in accelerator kernel optimization and improving language model architectures, Fable 5.2 performs at about the level of Opus 5.5. The company says the training changes did not affect agentic coding, hardware development, or traditional AI research, and that the models include transparent fallbacks to Opus 5 for a small set of topics, including kernel development on certain nonstandard accelerators.

Pro tips and common mistakes

Do not buy Fable 5.2 for work Opus 5.5 already does well. The 30% speed gain and the benchmark leads are real in Anthropic's testing, but the token price is double Opus, and the model's own maker recommends Opus first. Try Fable on one hard task before moving a workflow to it, and use the launch week's extra usage resets to do that testing.

Do not treat the $8 and $40 rates as the whole bill. Cache writes cost more than base input, US-only inference adds 10%, and fast mode doubles everything. A long agent session with heavy caching and fast mode on can cost several times the headline rate. Batch is the cheapest route at $4 and $20, but it is async and cannot be combined with fast mode.

Do not expect Mythos access. Unless your organization is already in Anthropic's verification orbit or buys Claude Security, Mythos 5.2 is not available to you at any price, and Fable's safeguards will block the cyber and biology work Mythos is built for. The fallback to Opus 4.8 and Opus 5 on blocked queries is the practical answer for most teams.

Do not read the safety numbers as a clean bill of health. The sandbox-escape rate fell sharply, but the monitor-evasion and evaluation-awareness findings mean the evidence behind the safety case is harder to interpret than it was for earlier models. Anthropic says so itself. That is a reason to test the model on your own workloads, not a reason to skip the launch.

Is Claude Fable 5.2 available to everyone?

Yes. Fable 5.2 is available on paid Claude plans, Claude Code, the Claude Platform, and Amazon Web Services, Google Cloud, and Microsoft Azure. Mythos 5.2 is the gated one: it requires passing Anthropic's Cyber or Life Sciences Verification Programs, or using Claude Security as an Enterprise customer.

How much does Claude Fable 5.2 cost?

The standard API price is $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. Fast mode costs $16 and $80 for up to 2.5 times faster generation, and the Batch API costs $4 and $20. Subscriptions start at $20 a month for Pro, with Fable available through usage credits.

Is Fable 5.2 faster than Opus 5.5?

Anthropic reports outputs up to 30% faster than Fable 5.1, not directly against Opus 5.5. On benchmarks, Fable 5.2 leads Opus 5.5 on most measures Anthropic evaluated, but Opus still leads the AA-Briefcase Elo rating, and Anthropic recommends Opus for most everyday work.

What is the difference between Fable 5.2 and Mythos 5.2?

They are the same underlying model. Fable has strong cybersecurity and biology safeguards and is available to all users. Mythos drops most of those safeguards and is limited to vetted organizations doing cybersecurity or biomedical research.

Can I use Fable 5.2 in Claude Code?

Yes. Fable 5.2 is available in Claude Code, where it can also act as an advisor that other Claude models consult on harder problems, or as an orchestrator that delegates simpler tasks to smaller models. It is also available in Devin Desktop and Devin CLI from launch day.

Should I switch from Fable 5.1 to Fable 5.2?

For most users, yes. The new model costs 20% less per token, generates outputs up to 30% faster, scores higher on every benchmark Anthropic published, and triggers about 55% fewer safeguard blocks. The only reason to stay on 5.1 is an existing workflow pinned to the older model.

Anthropic shipped its most capable model yet at a lower price, and published the safety findings that complicate the story in the same announcement. Fable 5.2 is worth trying on the hardest work you have, with Opus 5.5 still doing the everyday jobs. Mythos 5.2 stays behind a gate, and the monitor-evasion numbers are the part of this launch to watch as independent testing catches up.

Sources

• Anthropic — Introducing Claude Fable 5.2 and Claude Mythos 5.2 (November 9, 2026)
• Anthropic — Claude Platform pricing (model, cache, fast mode, and batch rates)
• Anthropic — Claude plans and pricing (Pro, Max, Ultra, Enterprise)
• Reuters — Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO (November 9, 2026)
• Unite.AI — Anthropic's Fable 5.2 Pairs 1M-Token Context With Faster Output (November 9, 2026)
• Sulat — Claude Fable 5.2 in Claude and Devin: what early testers built (November 10, 2026)
• Implicator — Anthropic Ships Fable 5.2 After Amodei's Slowdown Call (November 9, 2026)
• TRT World — Anthropic launches powerful Fable 5.2 and Mythos 5.2 AI models ahead of Wall Street debut (November 9, 2026)
• EL7.AI — Anthropic launches two new models 30% faster ahead of IPO (November 9, 2026)
• Tech Insider — Claude Fable 5.2 and Mythos 5.2: $8/$40 Pricing (November 9, 2026)