---
type: "Evidence Item"
title: "Claude Fable 5.2 and Mythos 5.2: frontier scores at a lower price, effort-dependent"
description: "Anthropic released Claude Fable 5.2 and Mythos 5.2 on 9 November 2026 as one underlying model under two safeguard regimes, priced at $8 per million input and $40 per million."
resource: "https://www.anthropic.com/claude-fable-and-mythos-5-2"
tags: ["appendix-iii", "vendor", "anthropic"]
timestamp: "2026-11-09"
category: "vendor"
publisher: "Anthropic"
cope_score: 76
confidence: 0.85
---

# Claude Fable 5.2 and Mythos 5.2: frontier scores at a lower price, effort-dependent

# Claim

Anthropic released Claude Fable 5.2 and Mythos 5.2 on 9 November 2026 as one underlying model under two safeguard regimes, priced at $8 per million input and $40 per million output tokens, 20% below Fable 5.1 at $10/$50, with outputs reported up to 30% faster. Vendor-reported benchmark figures, each with the effort level Anthropic states: FrontierCode 2.0 49.1% at Xhigh effort and 45.3% at High, against Opus 5.5 at 45.5% and GPT-6 Astra at 42.8%; FrontierSWE v2 69.9% against Astra 65.5% and Opus 5.5 62.3%; BenchCAD 2.1 with tools 83.5 at Max effort against Opus 5.5 at 65.3; Humanity’s Last Exam with tools 74.2%. Anthropic also reports High effort scoring about level with Opus 5.5 at Max effort for roughly half the cost per task, and about 55% fewer safeguard blocks than Fable 5.1. Independently of Anthropic, Vals AI ranks Fable 5.2 first of 36 models on its GDP-weighted Vals Index v2.2 at 71.31%, ahead of Claude Sonnet 5.5 at 70.85% and Gemini 4 Carbon at 70.45%.

# Relevance

Appendix III, sections one and two: model capability and vendor threshold evidence

# Oracle Verdict

Capability pressure rising while price falls is the combination that matters here: the vendor’s own claim is that High effort now buys Opus-at-Max performance for roughly half the cost per task, which lowers the cost of routing a task to a model rather than a person. The evidence does not run one way. Gemini 4 Carbon leads Fable 5.2 on AutomationBench business workflows at 53.9% against 51.4%; on AA-Briefcase v1.1 Fable 5.2 wins the rubric pass rate at 66.1% against Opus 5.5 at 61.2% but loses the head-to-head Elo at 1,785 against 1,800; and the Vals Index lead over Sonnet 5.5 is 0.46 points, inside noise. Every figure recorded here is a vendor benchmark or a third-party index, not measured labour displacement.

# Metadata

* Publisher: Anthropic
* Category: vendor
* Sector: General AI capability
* Capability: Agentic coding and professional task execution
* Cope score: 76
* Confidence: 0.85

# Related Concepts

* [Live evidence index](index.md)
* [Thesis](../thesis.md)

# Citations

[1] [Claude Fable 5.2 and Mythos 5.2: frontier scores at a lower price, effort-dependent](https://www.anthropic.com/claude-fable-and-mythos-5-2)