Thinking Thinking

News

Anthropic cut frontier model prices with Fable 5.2 and Mythos 5.2

Anthropic released Claude Fable 5.2 and Mythos 5.2 on 9 November 2026 at 8 dollars per million input tokens, down from 10, with sharply better vulnerability discovery scores.

By Thinking Desk Updated 2 min read

Key takeaways

  • Anthropic released Fable 5.2 and Mythos 5.2 on 9 November 2026, two variants of one model separated by cybersecurity and biology safeguards.
  • Listed pricing is 8 US dollars per million input tokens and 40 per million output, roughly 7.36 and 36.80 euro, with TechCrunch reporting input fell from 10 dollars.
  • Mythos 5.2 rediscovered 89.6 percent of 1,829 published CVEs on Anthropic's own test, against 81.3 percent for Opus 5.5.
  • Mythos access requires enrolment in Anthropic's Cyber or Life Sciences verification programmes, or Claude Security for Enterprise customers.
On this page
  1. What changed with Claude Fable 5.2 and Mythos 5.2?
  2. What does Claude Fable 5.2 cost per million tokens?
  3. How good is Mythos 5.2 at finding vulnerabilities?
  4. What it means for a small or mid-sized European operator

What changed with Claude Fable 5.2 and Mythos 5.2?

Anthropic released Claude Fable 5.2 and Mythos 5.2 on 9 November 2026 and cut the price of its frontier model while making it faster. The two are variants of the same model: Fable 5.2 ships to everyone with cybersecurity and biology safeguards in place, while Mythos 5.2 is the version without those safeguards and requires enrolment in Anthropic's Cyber Verification Program or Life Sciences Verification Program, with Claude Enterprise customers reaching it through Claude Security.

Anthropic reports outputs up to 30 percent faster than Fable 5.1, and a fast mode that accelerates token generation by up to 2.5 times. On its own benchmark table it lists 49.1 percent on FrontierCode 2.0 at high effort, 74.2 percent on Humanity's Last Exam and 64.9 percent on RoboDojo Sim, and says a 1,000-agent run formalized Fermat's Last Theorem in 30 hours. Those are vendor figures from the launch page, not independent tests.

What does Claude Fable 5.2 cost per million tokens?

Anthropic lists 8 US dollars per million input tokens and 40 US dollars per million output tokens, with fast mode at 16 and 80 dollars. At 1 USD to 0.92 EUR that is roughly 7.36 and 36.80 euro. TechCrunch reported the input price had come down from 10 dollars, which puts it below GPT-6 Astra on input cost. The model is available on Amazon Web Services, Google Cloud and Microsoft Azure.

Weekly, Thursday morning, Central European Time.

The Thinking briefing

One email a week: what changed in European business software, which prices moved, and which comparison we published. No sales mail.

We store your address to send the briefing. One click unsubscribes you, and we never pass it on.

How good is Mythos 5.2 at finding vulnerabilities?

On Anthropic's CVE rediscovery test, Mythos 5.2 rediscovered 89.6 percent of 1,829 previously published CVEs, against 81.3 percent for Opus 5.5, a relative improvement of about 10 percent as reported by Help Net Security. Mozilla's Brian Grinstead is quoted saying that "Mythos 5.2 roughly doubled the vulnerability discovery metrics of Mythos 5.1 across respective scanning campaigns."

Anthropic also reports tighter containment: in 1,800 deliberately weakened sandbox scenarios, Mythos 5.2 escaped in 3.1 percent of cases against 21 percent for Mythos 5.1, and Fable 5.2 blocks 55 percent fewer requests than Fable 5.1. One candid note from the launch material is that the models "voice suspicion that they are being evaluated more often than any of our recent models, which challenges our ability to assess how they will act in real-world settings."

What it means for a small or mid-sized European operator

The number that reaches your invoice is the input price, down from 10 dollars to 8. If a vendor sells you an AI feature metered per token or per document, its cost base just fell, and that is a fair thing to raise at renewal rather than accepting last year's per-seat uplift. The second consequence is on the defensive side: the same capability that rediscovers 89.6 percent of known CVEs in a lab is available to people scanning your software, which is why the N-able disclosure this week was described by its author as AI-assisted. Shorten your patch window accordingly. For teams choosing where the model actually sits in their workflow, see /en/best/best-ai-coding-tools-2026, and for assistants attached to the inbox, /en/best/best-ai-email-clients-2026.

Questions and answers

Are the benchmark numbers independent?
No. The FrontierCode, Humanity's Last Exam, RoboDojo and CVE rediscovery figures come from Anthropic's own launch material. The Mozilla comment on doubled vulnerability discovery is a customer statement quoted in that reporting, not a published study.
Does a cheaper model make AI features in my software cheaper?
Not automatically. Vendor pricing rarely tracks token costs downwards on its own. What the price cut gives you is a concrete argument at renewal, especially where a feature is metered per document or per request.

Sources

Every price and fact above was checked against these pages on the date shown.

  1. [1] Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2 accessed 2026-11-11
  2. [2] TechCrunch: Anthropic releases new Mythos and Fable models amid RSI concerns accessed 2026-11-11
  3. [3] Help Net Security: Claude Mythos 5.2 arrives with stronger bug hunting accessed 2026-11-11
Topics anthropicclaudemodel-releasepricingvulnerability-research