Anthropic Launches Fable 5.2 and Mythos 5.2 at $8/M Tokens

Executive Overview

Anthropic on Monday announced the launch of Claude Fable 5.2 and Claude Mythos 5.2, priced from $8 per million input tokens and $40 per million output tokens, with generation up to 30% faster and an optional Fast Mode that raises token generation speed up to 2.5x. The release follows a METR group report indicating that internal models accelerate Anthropic's own progress by roughly 2x, intensifying debate over recursive self-improvement (RSI). Fable 5.2 is available to all users, while Mythos 5.2 is gated behind verification programs for life sciences and cybersecurity researchers.

Deep-Dive Features & Architecture

Claude Fable 5.2 and Claude Mythos 5.2 are built on the same base model, but the core difference lies in the safeguards layer: Fable 5.2 carries strong cybersecurity and biosecurity safeguards that allow general availability, while Mythos 5.2 is available to institutions researching cybersecurity and biomedical applications through expanded verification programs. In preliminary tests, the two models helped Anthropic engineers complete software migrations that were practically impossible, design three working electronic devices from scratch, solve two open problems in black hole physics, make tangible progress on robotics evaluations, and formally prove Fermat's Last Theorem in 30 hours.

On efficiency, Fable 5.2 at Medium effort (the default in Claude and Claude Code apps) can complete complex tasks with fewer tokens than Fable 5.1. In early FrontierCode 2.0 tests, Fable 5.2 at High effort scored roughly equal to Opus 5.5 at maximum effort at about half the cost per task, while at Xhigh level it posted the highest score among all models offered. Fable 5.2 can also act as an Advisor consulted by Haiku, Sonnet, and Opus models, or as an Orchestrator delegating simpler tasks to smaller models.

On safety, Anthropic ran extensive pre-launch safety tests, including new evaluations for concerning behaviors such as sandbox escapes, and confirmed that high risks from misalignment remain low and that the model performed at or above modern models on most alignment metrics. The company notes that Mythos 5.2...

Benchmark & Competitive Performance

The published benchmark figures position Fable 5.2 as a substantial generational jump over Fable 5.1 and a direct challenger to Opus 5.5. On FrontierCode 2.0, Fable 5.2 reached 49.1% versus 32.4% for Fable 5.1 and 45.5% for Opus 5.5. It also posted 69.9% on FrontierSWE v2, 51.4% on AutomationBench, 89.3% on OSWorld 2.1, and 74.2% on Humanity's Last Exam with tools. On the security side, Mythos 5.2 recorded 76.3% on CyScenarioBench and 89.6% on CVE Rediscovery.

• Cost efficiency: High-effort Fable 5.2 matches max-effort Opus 5.5 at roughly half the cost per task.
• Speed: Up to 30% faster generation, with Fast Mode at up to 2.5x.
• Pricing edge: $8/$40 per million tokens undercuts the previous $10 input tier and the main competitor GPT-6 Astra.
• Security: Strongest protection to date against distillation attacks, plus new sandbox-escape evaluations.

Industry Impact & Enterprise Adoption

The launch lands in a crowded frontier race that includes GPT-6 Astra and Gemini 4 Carbon. Anthropic's decision to cut input pricing to $8 per million tokens while adding a premium Fast Mode at $16/$80 creates a two-tier cost structure aimed at both high-volume production workloads and latency-sensitive agentic pipelines. The availability of Fable 5.2 to all users, combined with ZDR support through Enterprise Frontier Safeguards, addresses a long-standing enterprise requirement and positions the model for regulated industries in the Middle East and globally. The METR finding that internal models accelerate Anthropic's progress by roughly 2x adds a governance dimension: the RSI debate is no longer theoretical, and Anthropic's staged access model — general availability for Fable 5.2, verification-gated access for Mythos 5.2 — is effectively a policy response to that risk.

Conclusion

Claude Fable 5.2 and Mythos 5.2 deliver a concrete package: $8/$40 per million token pricing, 30% faster generation, a 2.5x Fast Mode, and benchmark leadership on FrontierCode 2.0 at 49.1%. The unresolved items are the explicit context window figure, which Anthropic has not stated in the published excerpt and which is documented in the System Card, and the full Mythos 5.2 safety disclosure. For enterprise buyers, the combination of lower input pricing, ZDR compatibility, and advisor/orchestrator roles makes Fable 5.2 a credible default for 2026 agentic deployments.

Media Source: TechCrunch AI | البيان الرسمي للشركة: المصدر الأصلي | Fact Verification & Analysis: AI Tools Oasis