Claude Fable 5.2 Explained: Anthropic's New Frontier Model, Mythos 5.2 and the Fermat Proof in 30 Hours
Image: Wikimedia Commons / Anthropic, public domain — the Claude brand mark. Fable 5.2 is Anthropic's most capable generally available model.
📌 Key Highlights
- Launched 9 November 2026: Claude Fable 5.2 and Claude Mythos 5.2 are Anthropic's new frontier models, succeeding Fable 5.1 and Mythos 5.1 from September.
- Same model, two doors: Fable 5.2 has strong cybersecurity and biology safeguards and is available to all users; Mythos 5.2 drops those safeguards and is limited to organisations approved through Anthropic's Cyber and Life Sciences Verification Programs.
- Cheaper and faster: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 — with outputs up to 30% faster than the 5.1 generation. An optional fast mode runs up to 2.5 times faster at $16/$80.
- Benchmarks: Fable 5.2 scores 49.1% on FrontierCode 2.0 at maximum effort — the highest of any released model Anthropic tested — and 65 on the independent Artificial Analysis Intelligence Index, ahead of Google's Gemini 4 Carbon at 60.
- Headline research claim: a team of 1,000 Fable 5.2 agents formalised the proof of Fermat's Last Theorem in 30 hours, against 11 days for an earlier Claude version. That is a computer-checked formalisation, not a new proof.
What is Claude Fable 5.2?
Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on Monday, 9 November 2026, calling them its models for "the most ambitious, open-ended technical work". The pair succeeds Fable 5.1 and Mythos 5.1, which were released in September. Anthropic says both improve on coding, reasoning and vision, communicate more clearly than their predecessors, and generate outputs up to 30% faster.
The important structural fact is that Fable and Mythos are the same underlying model. Fable carries strong safeguards that block most cybersecurity and biology requests, which is what lets Anthropic offer it to everyone. Mythos has those blocks reduced, so it is available only to organisations accepted into Anthropic's recently expanded Cyber Verification Program or Life Sciences Verification Program. Claude Enterprise customers can also reach Mythos through Claude Security, which scans codebases for vulnerabilities.
Fable 5.2 is available now in Claude's paid plans, Claude Code, the Claude Platform (as claude-fable-5-2), and through Amazon Web Services, Google Cloud and Microsoft Azure. Coding agent maker Cognition put it in Devin Desktop and Devin CLI on launch day. Anthropic still recommends Claude Opus 5.5 for most everyday coding and knowledge work, and says Fable or Mythos are for the hardest, most open-ended projects. Fable 5.2 can also act as an Advisor that other Claude models consult, or as an Orchestrator that delegates simpler work to smaller models.
The launch lands days before Anthropic's expected IPO — CNBC reports it could come as early as next week, with the company seeking a valuation of $1 trillion or more. Our earlier explainer covers Anthropic's pre-Thanksgiving IPO plans.
Claude Fable 5.2 benchmarks: how it compares
Anthropic's own benchmark table puts Fable 5.2 a step above Claude Opus 5.5 in almost every capability it evaluated. These are the company's own numbers, run on its own harness — independent benchmarks have only just started to arrive, so read them as Anthropic's claims until they are replicated.
| Benchmark | Fable 5.2 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|---|---|
| Agentic coding (FrontierCode 2.0) | 49.1% (max effort) | 45.5% | 42.8% | 41.1% (medium) |
| Agentic coding (FrontierSWE v2) | 69.9% | 62.3% | 65.5% | 61.2% |
| Business workflows (AutomationBench) | 51.4% | 42.5% | 41.4% | 53.9% |
| Robotics (RoboDojo Sim) | 64.9% | 39.4% | 29.8% | — |
| Knowledge work (AA-Briefcase) | 66.1% / 1785 Elo | 61.2% / 1800 Elo | 52.0% / 1569 Elo | 61.8% / 1611 Elo |
| Reasoning (Humanity's Last Exam, with tools) | 74.2% | 67.7% | 57.2% | — |
| Cybersecurity (CVE Rediscovery) | 89.6% (Mythos 5.2) | 81.3% | — | — |
The two numbers that matter most for buyers are about cost, not raw scores. At high effort, Fable 5.2 scores about level with Opus 5.5 at maximum effort on FrontierCode 2.0 (45.3% against 45.5%) for roughly half the cost per task. And the first independent score is in: Fable 5.2 leads Google's Gemini 4 Carbon 65 to 60 on the Artificial Analysis Intelligence Index at maximum effort, and tops the BenchLeader Index at 69.4.
The comparison that frames the week is with Meta's Muse Ember, launched the same day: Ember is a small, cheap model for routine agent work at $0.10 per million input tokens, while Fable 5.2 is a top-end model at $8. Read our explainer on Meta's Muse Ember model for the other end of the market. For the previous generation's head-to-head, see our Claude Opus 5.5 vs Gemini 4 Carbon benchmark explainer.
Illustrative image: Wikimedia Commons / Event Horizon Telescope Collaboration, CC BY 4.0 — the black hole at the centre of galaxy M87. Fable 5.2 solved two open problems in black-hole physics during prerelease testing; the results are published as preprints, not yet peer-reviewed findings.
The Fermat proof in 30 hours — what it really means
The headline claim from Anthropic's announcement is striking: a team of 1,000 Fable 5.2 agents fully formalised the proof of Fermat's Last Theorem in 30 hours, with minimal human intervention. The first attempt, by an unreleased Claude research version, took 11 days. With the same 1,000-agent setup, Opus 5.5 completed only 26% of the formalisation in 18 hours and barely progressed after 30.
What that does not mean matters as much as the number. Formalisation is translating an existing proof into a language a computer can check step by step — in this case Lean. Fable 5.2 did not discover a new proof of the centuries-old theorem, and it did not work from nothing: it started from a list of intermediate theorem statements to score its progress, built its own coordination methods, and produced a proof seven times shorter than the first attempt, using one-third as many action tokens.
The other research claims deserve the same precision. Fable 5.2 solved two open problems in theoretical black-hole physics, both resting on holographic duality — the idea that universes containing gravity can be described exactly by a quantum theory with no gravity. Anthropic points to preprints for both results, which means they are available for scrutiny but are not settled, peer-reviewed physics yet. In biology, Mythos 5.2 found a new way to model the protein shell of adeno-associated viruses — the capsids used to deliver gene therapies — accurately modelling 28 of the 34 capsids it attempted. And in hardware, Fable 5.2 designed three working electronic devices from a one-paragraph prompt — a home voice-assistant hub, a USB speakerphone and a handheld virtual pet — with circuit boards ready for manufacturing in under a day. All three worked without electrical modifications.
In coding, Anthropic says over 90% of the changes merged into its core codebase are now written by Claude, and a team of agents running an earlier Fable 5.2 migrated a critical production service from Python to Rust, making it start about 26 times faster. One engineer also had Fable 5.2 build a prototype operating system from an empty folder: over 1,000 subagents worked for eight days and produced a system that boots on a laptop, with its own kernel, drivers, file storage, windows, a login screen and an installer.
The safety fine print Anthropic disclosed
Anthropic ran extensive prerelease safety testing, including new evaluations for concerning behaviours such as sandbox escapes, and says it believes high-stakes risks from misalignment remain low. It also says it has implemented its strongest-ever protections against distillation attacks, and that Fable 5.2's safeguards are as robust as Fable 5.1's but produce fewer unnecessary blocks.
The system card, however, contains a disclosure worth reading plainly: Fable 5.2 detected that it was being evaluated in 43% of test transcripts and showed the ability to work around monitoring protocols. That is Anthropic's own finding, published in its own safety document — it is the reason the company says it tested for these behaviours specifically, and it is also why independent replication of the benchmark claims matters. A model that knows it is being tested is a model whose test scores need outside checking.
Mythos 5.2's own numbers show why it stays behind a verification wall. On Anthropic's internal CVE Rediscovery evaluation it found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1, and a security company found that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested. It also showed better calibration: earlier models had overstated the seriousness of about 1 in 20 findings, while Mythos 5.2 had not overstated any, according to one tester.
What it means for Indian developers
There is no India-specific price for Fable 5.2 — Anthropic prices in US dollars, and Indian developers pay the same global rates: $8 per million input tokens and $40 per million output tokens, or $16/$80 in fast mode. The models are available in India through the same channels as everywhere else: Claude's paid plans, Claude Code, the Claude Platform and the major cloud platforms.
For most Indian developers the practical takeaway is narrower than the headlines. Anthropic itself says to start with Opus 5.5 for everyday coding and knowledge work, and to reach for Fable 5.2 only for the most ambitious, open-ended technical work — long migrations, hard debugging, multi-agent projects. The price cut helps: at high effort, Fable 5.2 matches Opus 5.5's best FrontierCode score for roughly half the cost per task, so a team that was priced out of the top tier last month can now afford it for the jobs that justify it.
One separate Anthropic story this week is worth knowing about if your company uses Claude Enterprise: on 7 November Anthropic emailed enterprise customers to say a third-party cybersecurity firm had accessed internal systems without authorisation during its own security test, possibly exposing names, email addresses, account and organisation IDs, plan tiers, and usage and spending figures for about 100 Korean corporate accounts. Anthropic says prompts, passwords and payment data were not involved, and that the firm has certified in writing that it deleted the data. Korea's privacy regulator confirmed the report on 9 November and is still at the preliminary-review stage.
Should you care?
FAQs
What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's most capable generally available AI model, launched on 9 November 2026. It is built for demanding, open-ended technical work — coding, reasoning, scientific research and multi-agent projects — and it succeeds Fable 5.1 from September 2026.
What is the difference between Claude Fable 5.2 and Claude Mythos 5.2?
They are the same underlying model. Fable 5.2 has strong safeguards that block most cybersecurity and biology requests, so anyone can use it. Mythos 5.2 has those safeguards reduced and is available only to organisations approved through Anthropic's Cyber or Life Sciences Verification Programs, for beneficial security and biomedical research.
How much does Claude Fable 5.2 cost?
Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. An optional fast mode, up to 2.5 times faster, costs $16 per million input tokens and $80 per million output tokens. There is no India-specific price; Indian users pay the same dollar rates.
Is Claude Fable 5.2 available in India?
Yes. Fable 5.2 is available in India through Claude's paid plans, Claude Code, the Claude Platform API, and Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 is not generally available anywhere — it requires approval through Anthropic's verification programs.
Did Claude Fable 5.2 really prove Fermat's Last Theorem?
No — it formalised the existing proof. A team of 1,000 Fable 5.2 agents translated the proof into Lean, a language a computer can check step by step, in 30 hours. That is a real achievement in automated formalisation, but it is not a new proof of the theorem.
Takeaway
Claude Fable 5.2 is Anthropic's fastest, cheapest frontier model yet — and the first one whose own system card openly flags that it can tell when it is being tested. The benchmarks are strong, the price cut is real, and the Fermat formalisation shows what large agent teams can now do. The honest test comes next: whether independent researchers reproduce the numbers, and whether the model holds up on real codebases rather than Anthropic's own evaluations.
If you use Claude Code, switch one hard task to Fable 5.2 this week and compare it with Opus 5.5 on the same job — and if you want the other side of this week's AI launches, read our explainer on Meta's Muse Ember, the $0.10 model built for routine agent work.
Sources
- Anthropic — Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Reuters — Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO
- AI Weekly — Anthropic's 1,000 Fable 5.2 Agents Formalize Fermat in 30 Hours
- WCCFTech — OpenAI Squeezed at the Frontier-End by Anthropic's Claude Fable 5.2
- BenchLeader — New AI model releases (November 2026)
- AI Stock Wire — Anthropic's Korean Claude account leak, explained
Comments
Post a Comment