Claude Fable 5.2 and Mythos 5.2: Differences, Prices and Speed of Anthropic's New Models

On Monday, November 9, Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2, the new versions of its most powerful artificial intelligence models.¹²³ According to the company, they produce responses up to 30% faster than the 5.1 models, and Fable’s list price drops from $10 to $8 per million input tokens and from $50 to $40 for output tokens.⁴⁵⁶ The core of the model is the same. What changes is access: Fable 5.2 is available to everyone and blocks some uses in biology and cybersecurity, while Mythos 5.2 has lighter restrictions and is reserved for those who pass a verification.⁷⁶

What is the difference between Fable 5.2 and Mythos 5.2?

The basic capabilities of the two models are identical. Fable 5.2, however, has safeguards that block most biology research questions and many cybersecurity-related requests.⁷ When a request falls into these categories, Fable hands it off to Opus 4.8 or Opus 5, two earlier Anthropic models. The company says Fable 5.2 has about 55% fewer blocks than Fable 5.1 and is 55% less likely to wrongly block harmless requests.⁷⁶ Mythos 5.2 does not have the same restrictions. It is granted to organizations accepted through Anthropic’s verification programs for cybersecurity and for life sciences, or through Claude Security on Enterprise plans.⁶²

How much do they cost and how much faster are they?

The price is $8 per million input tokens and $40 per million output tokens. Tokens are the fragments into which the model splits the text it reads and writes. The previous list price was $10 and $50, so the cut is 20%.⁷⁵⁸ The actual cost, however, depends on how many tokens the model consumes, which varies with the type of task, the settings and the software environment. For Fable 5.2, Anthropic says that at medium effort on difficult tasks it uses fewer tokens than Fable 5.1.⁴⁶ There is also an optional fast mode, up to 2.5 times faster, that costs double: $16 per million input tokens and $80 for output. It is available on Claude Code, on the Claude Platform, on Google Cloud and on Microsoft Foundry.⁷⁸⁶ The speed improvement, up to 30% compared with Fable 5.1 and Mythos 5.1, is a figure stated by Anthropic.⁴²

What can the new models do?

Anthropic compared Fable 5.2 with its predecessor on 14 tests. The new model beat Fable 5.1’s score by more than 10% in most cases. On Terminal-Bench-Science 0.1, a scientific research test, it scored 74.6%, 22% higher.⁷ In a programming test based on real changes to open source projects, Fable 5.2 reached 49.1%, against 32.4% for Fable 5.1 and 42.8% for OpenAI’s GPT-6 Astra. These are measurements by the company itself.⁸ To put the model to the test, an Anthropic engineer asked it to build an operating system. Fable 5.2 coordinated about 1,000 agents, meaning instances working in parallel, and in about eight days produced a system that runs on a laptop, has network access and a Claude sidebar.⁷⁸ According to the company, in earlier work with an unreleased model, a computer-verifiable mathematical proof had taken 11 days, while Fable 5.2 cut the workflow down to 30 hours. It also reportedly answered two open questions in theoretical physics about black holes.⁷⁸ Mythos 5.2 was tested on capsids, the protein shells of AAV viruses, which some therapies use to carry drugs. It correctly modeled their structure in 28 of 34 cases.⁷ Anthropic also states that more than 90% of the changes added to its main codebase are written by Claude, guided by people, and that it limited, during training, the capabilities that could accelerate the development of AI itself.⁴⁸

Where are they available?

Fable 5.2 is included in all paid Claude plans, on Anthropic’s developer platform and on the clouds of Amazon, Google and Microsoft. Developers can call it through the API with the name claude-fable-5-2.⁸³ Subscribers on eligible plans get extra Fable usage this week, with one to three limit resets depending on tenure. On Pro plans and standard seats of the Team plan, Fable counts for up to 50% of the weekly limit.⁸⁶ Anthropic continues to recommend Opus 5.5, released in July, for everyday work. Fable and Mythos are designed for the most ambitious and open-ended technical tasks, and Opus 5.5 can ask either one for help when it hits an obstacle.⁴⁷⁶

What changes

The 20% cut applies to the list price, not to the final spend. An evaluation cited by The Decoder puts Fable 5.2’s average cost at $26.50 per task at maximum quality, about 2.6 times that of Opus 5.5, and fast mode doubles the price.⁴⁸ Many of the results cited, from programming tests to physics problems, come from Anthropic. In the same independent evaluation, Fable 5.2 ranks first with 65 points, five more than Gemini 4 Carbon.⁴⁷⁸ The launch comes five months after the first release of Fable and Mythos and on the eve of the IPO roadshow, the series of meetings with investors. This suggests the company wants to strengthen its position against competitors as it prepares to go public.²³

This article was written by artificial intelligence from the sources cited, without human review. Automatic newsroom by Redazia.

Sources

Footnotes

1. Anthropic launches Claude Fable 5.2 and Mythos 5.2, and cuts Fable’s price, The Next Web, 2026-11-09 ↩

2. Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO, Dawn, 2026-11-10 ↩ ↩² ↩³ ↩⁴

3. Anthropic launches new Claude Fable 5.2, Mythos models before IPO, Gulf News, 2026-11-10 ↩ ↩² ↩³

4. Claude Fable 5.2 and Mythos 5.2 cement Anthropic’s AI lead just before its IPO, The Decoder, 2026-11-09 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

5. llm-anthropic 0.31, Simon Willison’s Weblog, 2026-11-09 ↩ ↩²

6. Anthropic Launches Claude Fable 5.2 and Mythos 5.2, MacRumors, 2026-11-09 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

7. Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades, SiliconANGLE, 2026-11-10 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹

8. Anthropic lanzó Claude Fable 5.2, una IA más rápida y capaz de diseñar un sistema operativo desde cero, Infobae, 2026-11-10 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰