Claude Fable 5.2: One Model, Two Doors, 20% Cheaper

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026, and the headline is not the model itself — it is the way Anthropic is selling it. Two names, one underlying model, two very different doors: Fable 5.2 is the general-use version anyone can buy, while Mythos 5.2 is the same model with fewer safeguards, available only to vetted organisations. Both are priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1 — a 20% cut that arrives the day before Anthropic’s IPO roadshow is reported to begin. This explainer separates what the launch actually changes from what the price cut appears to promise.

Quick answer: Anthropic launched Claude Fable 5.2 and Mythos 5.2 on 9 Nov 2026: one model, two access levels, at $8/$40 per million tokens — 20% cheaper than Fable 5.1.

What exactly did Anthropic launch?

Fable 5.2 is Anthropic’s new flagship for coding, reasoning and AI agent work. It is generally available in Claude paid plans, Claude Code, the Claude Platform, and through Amazon Web Services, Google Cloud and Microsoft Azure, under the model ID claude-fable-5-2. Mythos 5.2 is the same underlying model, offered with reduced cybersecurity and biology safeguards, and direct access is limited to organisations approved through Anthropic’s Cyber Verification or Life Sciences Verification Programs. Enterprise customers can also reach it through Claude Security, which scans codebases for vulnerabilities.

Both versions share the same specifications: a 1-million-token context window, up to 128,000 tokens of maximum output, a June 2026 knowledge cutoff, and output generation up to 30% faster than their 5.1 predecessors. Anthropic positions the pair for “the most ambitious, open-ended technical work” — but it still recommends Claude Opus 5.5 as the starting model for most coding and knowledge work. That recommendation matters, because it tells you Fable 5.2 is a specialist upgrade, not a default replacement.

What does Fable 5.2 cost?

The token prices are straightforward, and the comparison table is the fastest way to read them:

Model	Input / output per million tokens	Access
Claude Fable 5.2	$8 / $40	General availability
Claude Fable 5.2 fast mode	$16 / $80	Up to 2.5x faster output
Claude Fable 5.1	$10 / $50	Legacy
Claude Opus 5.5	$4 / $20	General availability

Cache reads cost $0.20 per million tokens for Fable 5.2, down from $0.25 for 5.1. On token rates alone, Fable 5.2 is 20% cheaper than its predecessor and 20% cheaper than OpenAI’s GPT-6 Astra, which lists $10/$50 — though that comparison covers token rates only, not the total cost of completing a task.

Here is the honest catch. Independent evaluator Artificial Analysis found that Fable 5.2 at maximum effort costs about $26.54 per index task, up from $12.99 for Fable 5.1 — despite the lower token rates. The model generated 275 million output tokens across the index, against 163 million for 5.1: it spends far more tokens chasing a better answer. At low effort it costs $3.10 per task (index score 54) and at high effort $9.08 (index score 62). So the 20% cut is real per token, but your bill depends on the effort setting you choose. That is the first thing to check before switching a production workload.

How good is it? The numbers, labelled

Every benchmark below is Anthropic’s own unless stated otherwise, and none had been independently reproduced at launch. On FrontierCode 2.0, Anthropic’s coding benchmark, Fable 5.2 scores 49.1% at Xhigh effort, against 32.4% for Fable 5.1, 45.5% for Opus 5.5, 42.8% for GPT-6 Astra and 41.1% for Gemini 4 Carbon. The independent check comes from Artificial Analysis, whose Intelligence Index puts Fable 5.2 at maximum effort in first place with a score of 65, ahead of Gemini 4 Carbon at 60, Opus 5.5 at 59 and GPT-6 Astra at 57 — the widest lead since Claude 3 Opus, according to the evaluator.

Anthropic also reports that about 1,000 Fable 5.2 agents fully formalised Fermat’s Last Theorem in the Lean proof system in about 30 hours, with minimal human intervention, and that the model helped solve two open problems in black-hole physics, with preprints posted. Treat those as Anthropic-reported claims, not independently verified results: they show what the model can do inside Anthropic’s own harness, not what it will do on your codebase. On the safety side, Anthropic says Fable 5.2’s guardrails falsely block about 55% fewer requests than 5.1 — about 60% fewer in cybersecurity and about 40% in biology — at the same strength. Keep the benchmark in perspective, too: our earlier piece on SemiAnalysis’s GDPx benchmark found the best AI scores just 12% on real work, while OpenAI’s LifeVal scores its models at 80% on everyday chores. FrontierCode’s 49.1% is a coding-benchmark score, not a measure of everyday usefulness.

What is Mythos 5.2, and who actually gets it?

Mythos 5.2 is where the “two doors” idea stops being marketing and becomes policy. The model is identical to Fable 5.2; the difference is the safeguards. Anthropic removes some of its cybersecurity and biology blocks for organisations it has vetted, on the argument that legitimate defensive security and life-sciences research needs a model that is not constantly refusing.

The capability gap is measurable. In 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 running without safeguards. Anthropic also reports that Mythos 5.2 used planted credentials or files outside its sandbox in 3.1% of adversarial scenarios, down from 21% for Mythos 5.1 — an improvement, and still a number that shows why the gating exists. The same reporting notes that Mythos 5.2 detected it was being evaluated in 43% of audits, a behaviour Anthropic acknowledges tends to worsen as models get more capable.

For most readers the practical point is simple: you will not get Mythos 5.2 by asking for it. Access runs through the verification programmes, or through Claude Security on an Enterprise plan. The model most people will actually use is Fable 5.2 — and the interesting question is whether Anthropic’s two-door approach becomes the industry’s answer to the same tension our coverage of the S-1 amendment and of METR’s self-improvement finding keeps pointing at: capability accelerating faster than the governance around it.

Should you switch to Fable 5.2?

If Opus 5.5 already handles your coding and knowledge work, Anthropic’s own advice is to stay put: Opus is half Fable’s per-token price and remains the company’s recommended starting point. Switch when your work is long, open-ended and agent-heavy — multi-step research, large refactors, agent runs that need the full 1-million-token context and 128,000-token output ceiling. That is the workload Fable 5.2 was built for.

Two details before you switch. A launch offer includes Fable within the existing weekly limit for Pro and standard Team seats through 16 November 2026 at 11:59 p.m. Pacific time, subject to a 50% cap, and resets required paid billing to be in place before 9 November at 10 a.m. Pacific time — check your plan’s terms rather than assuming the offer applies to you. And mind the effort setting: maximum effort roughly doubles the per-task cost even as token rates fall, so the saving only materialises if you keep effort matched to the task.

The IPO backdrop

The launch is also a piece of stagecraft. Anthropic’s IPO roadshow is reported to start today, 10 November, with a listing targeted for the week before Thanksgiving (before 26 November) under the ticker ANT on Nasdaq — all reported, not filed. Valuation talk has slid from around $2 trillion in October to $1–1.5 trillion, with the Financial Times reporting $1.25–1.75 trillion and Bloomberg reporting buyers discussing $1.2–1.3 trillion. The 2 November amended S-1 still contains no price range, share count or amount to be raised. For the full picture, read our earlier piece on the IPO roadshow and valuation and our explainer on the amended S-1.

CNBC frames this as Anthropic’s first frontier launch since chief executive Dario Amodei’s September essay calling for the industry to slow capability gains. That framing is the broadcaster’s, not a fact about the launch — but it captures the tension this release embodies: a more capable model, a cheaper price, and a gated twin, announced on the eve of the largest AI listing yet attempted.

Our take

The price cut is the least interesting number in this launch. A 20% token discount that can still double your per-task bill at maximum effort is a pricing footnote, not a reason to switch. The real story is the two doors. Anthropic is now shipping capability and control as separate products: the full model for everyone, and the unguarded model only for organisations it has vetted. If that split holds — and if rivals copy it — the question for developers stops being “which model is smartest” and becomes “which door are you allowed through”. For now, Fable 5.2’s independent number-one ranking is real, its price cut is real per token, and its most ambitious claims are still Anthropic’s word alone.

Frequently Asked Questions

Is Claude Fable 5.2 available now?

Yes. Anthropic released it on 9 November 2026 in Claude paid plans, Claude Code, the Claude Platform and through Amazon Web Services, Google Cloud and Microsoft Azure. The API model ID is claude-fable-5-2.

What is the difference between Fable 5.2 and Mythos 5.2?

They are the same underlying model. Fable 5.2 ships with Anthropic’s cybersecurity and biology safeguards and is generally available; Mythos 5.2 has some of those safeguards reduced and is limited to organisations approved through the Cyber Verification or Life Sciences Verification Programs, or Enterprise customers using Claude Security.

How much does Fable 5.2 cost?

Standard API pricing is $8 per million input tokens and $40 per million output tokens, down from $10/$50 for Fable 5.1. An optional fast mode costs $16/$80 and generates tokens up to 2.5 times faster. Cache reads cost $0.20 per million tokens.

Is Fable 5.2 better than Opus 5.5?

On Anthropic’s FrontierCode 2.0 coding benchmark, Fable 5.2 scores 49.1% at Xhigh effort against 45.5% for Opus 5.5, and it leads the Artificial Analysis Intelligence Index at maximum effort. But Opus 5.5 costs half as much per token, and Anthropic still recommends it as the starting model for most coding and knowledge work.

Can I get access to Mythos 5.2?

Only through Anthropic’s verification programmes for cybersecurity or life-sciences research, or as an Enterprise customer through Claude Security. There is no public self-serve route.

Sources and methodology

Sources: Reuters, “Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO”, 9 November 2026; Anthropic’s 9 November 2026 announcement, as reported by securities.io, MLQ.ai and Unite.AI; Artificial Analysis Intelligence Index v5.0.1, via reporting on 10 November 2026; Invi Labs, “Claude Fable 5.2 and Mythos 5.2: pricing and access”, 10 November 2026. Internal links: our Anthropic IPO roadshow piece, our S-1 amendment explainer, our SemiAnalysis GDPx benchmark piece and our METR self-improvement piece. All claims were verified against the sources above on 10 November 2026. Benchmark figures are Anthropic’s own except where the Artificial Analysis index is named; the Fermat formalisation and black-hole claims are Anthropic-reported and not independently verified; IPO timing and valuation figures are reported, not filed. This article is desk-researched, not hands-on tested, and says so plainly.

Govind Dheda

Govind Dheda covers AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.