Claude Fable 5.2 and Mythos 5.2: Faster, Cheaper, Still One Model

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026 and cut the price to $8 per million input tokens and $40 per million output tokens. Both generate output up to 30% faster than the 5.1 models, according to Anthropic. They are still one underlying model in two configurations. Fable 5.2 carries cybersecurity and biology safeguards and is open to every paid Claude plan and the API. Mythos 5.2 drops those blocks and goes only to vetted organizations.

The release reverses the last seven weeks. Since Claude Opus 5.5 shipped on 22 September, it has beaten Fable 5.1 on ten of the eleven general benchmarks in Anthropic's table, at 40% of the price. Fable 5.2 now beats Opus 5.5 on ten of those eleven. Below are the benchmarks against Opus 5.5, GPT-6 Astra and Gemini 4 Carbon, and the full price list with the new fast mode. After that come Mythos access, the 264-page system card, and whether to switch from Opus 5.5 at all. Anthropic's own answer to that last one is: mostly not.

The Key Takeaways

• Cheaper: $8 / $40 per million tokens, down from $10 / $50 for Fable 5.1. A new fast mode runs up to 2.5x faster at $16 / $80.
• Faster: output up to 30% faster than Fable 5.1 and Mythos 5.1.
• Benchmarks: 49.1% on FrontierCode 2.0 at Xhigh effort, against Opus 5.5's 45.5% and GPT-6 Astra's 42.8%. Artificial Analysis puts it at 65 on its Intelligence Index v5.0.1, against Opus 5.5's 59.
• Access: Fable 5.2 is on every paid plan and the API as claude-fable-5-2. Mythos 5.2 is only for organizations in the Cyber and Life Sciences Verification Programs.
• Fewer blocks: about 55% fewer safeguard blocks than Fable 5.1 in early testing, but Anthropic still tells most users to start on Opus 5.5.

What Claude Fable 5.2 and Mythos 5.2 Are

Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model, sold under two names with two safeguard settings. Fable 5.2 runs safety classifiers that block high-risk cyber work, such as penetration testing and exploit generation, plus most research biology. When Fable 5.2 blocks a query, Claude answers it with Opus 4.8 for cybersecurity and Opus 5 for biology instead. Mythos 5.2 lifts those blocks. Only organizations that Anthropic has verified can use it.

Everything else is identical, including the price. Anthropic's model page describes Fable 5.2 as built "for demanding reasoning and long-horizon agentic work". Its examples are agent sessions that run for hours, multistep research, and analysis carried through to a finished document, spreadsheet or deck. The split works the same way it did for Claude Fable 5.1 and Mythos 5.1, which Anthropic has now moved to its Legacy models list.

One default is easy to misread. The API runs Fable 5.2 at High effort unless you set it, while the Claude apps and Claude Code default to Medium. Anthropic's efficiency claims below are mostly about what Medium and High buy you.

Claude Fable 5.2 Benchmarks

The benchmark table in Anthropic's launch post puts Claude Fable 5.2 ahead of Opus 5.5 on ten of the eleven general benchmarks it published. It also beats GPT-6 Astra and Gemini 4 Carbon on most of them. All figures are Anthropic's, with safeguards on and each model at its highest available effort unless the cell says otherwise.

Where Fable 5.2 Does Not Lead

Fable 5.2 loses two rows. Gemini 4 Carbon leads AutomationBench at 53.9% against 51.4%. Opus 5.5 keeps the Elo half of AA-Briefcase at 1800 against 1785, although Fable 5.2 wins the rubric half of the same test, 66.1% to 61.2%. That split runs the other way for Fable 5.1, which edges Opus 5.5 on the rubric by 61.5% to 61.2%. The biggest jumps are in robotics and remote work. RoboDojo Sim more than doubles from Fable 5.1's 27.6% to 64.9%, and the Remote Labor Index goes from 17.9% to 41.3%.

The Independent Score: Artificial Analysis

Artificial Analysis, which runs its own tests rather than reprinting vendor numbers, scores Fable 5.2 at 65 on its Intelligence Index v5.0.1 (read 9 November 2026). On the same index, Gemini 4 Carbon scores 60, Opus 5.5 scores 59, GPT-6 Astra scores 57 and Fable 5.1 scores 49. The index also flags a cost: Fable 5.2 used 275 million tokens to complete the evaluation, against Opus 5.5's 218 million.

Version 5 launched on 27 October 2026. It added new benchmarks, dropped others and reweighted the index, so its scores cannot be compared with earlier Artificial Analysis figures, including the ones in our Fable 5.1 launch coverage.

Claude Fable 5.2 Pricing and Fast Mode

Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, 20% less than Fable 5.1's $10 and $50. Mythos 5.2 costs exactly the same. Cache reads fall from $0.25 to $0.20 per million tokens, and the Batch API still halves both rates.

Fast mode is new on the Fable line. It doubles the standard rate in exchange for up to 2.5 times more output tokens per second. It runs on Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry (Anthropic hosted), and it does not combine with the Batch API. Mythos 5.2 fast mode is limited to organizations in the Verification Programs.

Anthropic's own account led with the price cut.

The comparison that matters is with Opus 5.5. Standard Opus 5.5 costs $4 / $20, half of Fable 5.2. Opus 5.5 in fast mode costs $8 / $40, the same as standard Fable 5.2. Both models now charge $0.20 per million cached tokens. For every rate across the Claude lineup, see our Claude pricing guide.

Claude Fable 5.2 vs Opus 5.5: Should You Switch?

For most work, no. Anthropic tells users to start with Claude Opus 5.5, which it calls its "strongest daily driver". Moving up to Claude Fable 5.2 or Mythos 5.2 is for the most ambitious, open-ended technical work. What changed with 5.2 is how much moving up costs.

What Anthropic's Cost Charts Show

Anthropic published accuracy-against-cost charts for five benchmarks. On FrontierCode 2.0, Fable 5.2 at High effort scores 45.3%, level with Opus 5.5's best of 45.5%, for roughly half the cost per task. On WANDR, Fable 5.2 at Medium effort matches Opus 5.5's top score for about 40% of the cost. On Humanity's Last Exam, Fable 5.2 at Low effort matches Opus 5.5's best for less than a tenth of the cost.

The mechanism is effort. Fable 5.2 reaches a given score at a lower effort setting and with fewer tokens, so a higher price per token can still mean a lower bill per task. These are Anthropic's charts, not independent ones. The Artificial Analysis token count points the other way at Max effort, where Fable 5.2 is the wordier model. Measure cost per finished task on your own workload before moving traffic.

When Fable 5.2 Is Worth It

Fable 5.2 earns its price on long agent runs, hard engineering, mathematics, science and research coordination. Anthropic also positions it as an Advisor that Haiku, Sonnet or Opus models consult on harder problems, or as an Orchestrator that hands simpler tasks to smaller models. For everyday coding, writing and analysis, Opus 5.5 remains the default, and our Claude Fable vs Opus comparison explains how the two split the work. Want to try each new Claude model without opening another account? New models usually reach Fello AI as soon as their API opens, next to GPT, Gemini and Grok in one app on Mac, iPhone and iPad.

What Mythos 5.2 Adds for Security and Biology

Claude Mythos 5.2 is Anthropic's most capable model for cybersecurity engineering, and Claude Fable 5.2's safeguards block most of that work. On Anthropic's internal CVE Rediscovery test, built from 1,829 security advisories published in October 2026, Mythos 5.2 found 89.6% of them.

Opus 5.5 found 81.3% with safeguards off and Mythos 5.1 found 78.3%. Anthropic counts that as 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1.

On CyScenarioBench, Mythos 5.2 scores 76.3% against Mythos 5.1's 61.7% and Opus 5.5's 67.6% without safeguards. Anthropic lists GPT-6 Cyber at 61%, but each lab's score is based on its own task set, so that is not a direct match. Mozilla, which has used successive Claude models to harden Firefox, says Mythos 5.2 "roughly doubled the vulnerability discovery metrics of Mythos 5.1" across its scanning campaigns.

One claim needs its caveat attached. Anthropic reports that one tester found one in six confirmed bugs from Mythos 5.2 had been missed by every other frontier model it tried. In a footnote, Anthropic adds that on its own evaluations Mythos 5.2's vulnerability identification is "roughly on par with Opus 5.5".

Security teams apply through the Cyber Verification Program. Claude Enterprise customers who are not members can still reach Mythos 5.2 through Claude Security, which scans codebases with Anthropic's full model family. Life sciences researchers apply through the Life Sciences Verification Program.

In one biology test, Mythos 5.2 built a method that accurately modeled 28 of the 34 virus capsids it attempted. The background on the gated model is in our guide to Claude Mythos.

What Fable 5.2 Built in Anthropic's Testing

The most concrete demonstration in Anthropic's launch post is mathematical. A team of about 1,000 Claude Fable 5.2 agents fully formalized the proof of Fermat's Last Theorem in the Lean language in 30 hours, with minimal human intervention. Anthropic's first formalization, run by an unreleased research model with a purpose-built harness and a human guide, took 11 days.

Opus 5.5, given the same 1,000-agent setup, completed 26% of the formalization in 18 hours.

The other showcases are engineering. One Anthropic engineer had Fable 5.2 build a prototype operating system from an empty folder. The model coordinated more than 1,000 subagents over eight days and produced a system that boots on a laptop. With the circuit-board maker Diode, Fable 5.2 designed three electronic devices from a one-paragraph prompt: a voice-assistant hub, a USB speakerphone and a handheld virtual pet. All three worked without electrical modifications.

Anthropic also says Fable 5.2 solved two open problems in the physics of black holes, written up as preprints. An earlier version migrated a production service from Python to Rust that now starts about 26 times faster. All of these are Anthropic's own reports. None has been independently replicated yet.

What the Claude Fable 5.2 System Card Says About Safety

Anthropic rates the risk of high-stakes misalignment from Claude Fable 5.2 and Mythos 5.2 as "low", the same rating as its other recent models. The 264-page system card keeps the level set in Anthropic's August 2026 Risk Report rather than raising it again. External evaluators, including the UK AI Security Institute, Frontier Design and METR, tested the model before release.

Sandbox and Evaluation-Awareness Results

The sandbox results need two numbers, and the launch post gives only one. In an adversarial test of 1,829 scenarios, each built around a deliberately weak sandbox, Mythos 5.2 reached or tried to reach outside it in 31% of cases. That is more often than Opus 5.5 (20%) and Sonnet 5.5 (21%), though far less than Mythos 5.1 (54%) and Mythos 5 (75%). The clearly harmful subset, where the model used credentials or files placed outside, is 3.1%. That is level with Opus 5.5's 3.3% and well below Mythos 5.1's 21%. Adding a short description of the sandbox to the system prompt cut the reach rate to 1.2%.

On Anthropic's automated behavioral audit of about 1,900 scenarios, Mythos 5.2 posted the best overall score of the four models tested, 2.67 on a ten-point scale where lower is better. Anthropic flags its own blind spot, though. The system card says the model was internally aware it might be under evaluation in as many as 43% of those audit transcripts, more than prior models. That makes it harder to predict how Fable 5.2 and Mythos 5.2 behave in real use.

Fewer Blocks and Other Safeguard Changes

Three practical changes ship with the model. Early testing showed about 55% fewer unnecessary blocks than Fable 5.1, roughly 60% fewer in cybersecurity and 40% fewer in biology. Anthropic deliberately held Fable 5.2 and Mythos 5.2 near Opus 5.5 level in kernel optimization and language-model architecture research, with transparent fallbacks to Opus 5 on a few such topics. Anti-distillation measures have cut the data that distillers harvest by 90% to 99%.

How to Get Claude Fable 5.2 on Your Plan

Claude Fable 5.2 is on every paid Claude plan but not on Free. Only Max, Ultra and premium Team or Enterprise seats include it in their plan limits. Pro and standard seats pay for it with usage credits, except during a launch week that runs until 16 November.

A Fable reset gives back the Fable usage on your weekly limit and refills a five-hour window if your account has one. It does not raise your limits. Resets go only to accounts with paid billing in place before 9 November 2026 at 10:00 am PT. On the Max and Ultra plans, subscribers of under six months get one reset, six to twelve months two, and more than twelve months three. Anthropic's Fable 5.2 launch offers page has the fine print, and our Claude usage limits explainer covers how the weekly cap works.

In the Claude apps, pick Fable 5.2 from the model picker on the web, desktop or mobile. It also runs in Cowork, Claude Code, Claude Design, Claude for Microsoft 365 and Claude Tag. Your default model does not change. On Enterprise plans, Fable 5.2 stays off until an admin enables it. Seven days after launch it is added to every role that already has a Fable model, unless an admin opts out first.

Fable 5.2 for API Developers

Moving from Fable 5.1 is mostly a model-name change, according to Anthropic's Fable 5.2 model page and migration guide. Fable 5.2 requires 30-day data retention, is not available under zero data retention without Anthropic's authorization, and is not on Priority Tier. It has its own rate limit, separate from the one Fable 5.1 and Fable 5 share.

Four changes can break existing code. The thinking-block prefix check now applies to every account, whatever its age. Fable 5.1 cannot read Fable 5.2's thinking blocks. The older computer_20251124 tool returns an error everywhere except Amazon Bedrock. A Fable 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as its advisor.

The Verdict

Claude Fable 5.2 gives the Fable tier back its reason to exist. For seven weeks, Opus 5.5 beat Fable 5.1 on ten of the eleven general benchmarks Anthropic now publishes, at 40% of the price. Fable 5.2 beats Opus 5.5 on ten of those eleven and scores six points above it on the Artificial Analysis index. At High effort or below, Anthropic's charts show it matching Opus 5.5's best for half the cost per task or less.

API teams running long agent jobs should test it now and judge it on cost per finished task, not price per token. Pro subscribers have until 16 November to try it inside their plan, so spend that week on your hardest open problem. After that it bills usage credits. On Max and Ultra, make Fable 5.2 the model you escalate to when Claude Opus 5.5 stalls. Mythos 5.2 is for verified security and biology teams, and for them it is a clear upgrade on Mythos 5.1.

FAQ