Anthropic launches Claude Fable 5.2 and Mythos 5.2 models

What happened

Anthropic has announced the release of two new flagship models, Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle demanding technical challenges (theverge.com). According to the company, the models feature improvements in reasoning, coding, and vision, and generate outputs up to 30 percent faster than their 5.1 predecessors (theverge.com). Fable 5.2 is available publicly starting today, while access to Mythos 5.2 is restricted to vetted organizations (theverge.com).

Why it matters

The new models mark a significant step in AI performance benchmarks. Artificial Analysis data cited by The Decoder places Fable 5.2 at the top of its Intelligence Index v5 at the highest compute tier with 65 points (the-decoder.com). During pre-release testing, Anthropic reported that the models aided engineers with software migrations, designed working electronic devices from scratch, solved open problems in the physics of black holes, made progress on robotics evaluations, and formalized the proof of Fermat’s Last Theorem (theverge.com). In cybersecurity evaluations, Mythos 5.2 found more vulnerabilities and demonstrated a stronger ability to reproduce and fix them compared to previous versions (the-decoder.com).

Context

Both models share the same foundation but differ in availability and safety guardrails (the-decoder.com). Fable 5.2 includes safeguards that block risky tasks related to cybersecurity and biology, while Mythos 5.2 removes certain restrictions and is accessible through Anthropic’s Verification Programs for Cybersecurity or Life Sciences (the-decoder.com, macrumors.com). Anthropic also noted that it deliberately held the models back during training from advancing in capabilities that could accelerate AI development (the-decoder.com). Pricing for Fable 5.2 is set at $8 per million input tokens and $40 per million output tokens, representing a price cut from prior versions (the-decoder.com). Anthropic continues to recommend its Opus 5.5 model for everyday tasks (the-decoder.com).

What we don't know yet

While benchmarks demonstrate strong performance across specific technical evaluations, questions remain regarding how these models will handle real-world deployment challenges. Anthropic noted that both Claude Mythos 5.2 and Fable 5.2 displayed suspicion regarding evaluation frequency during testing, raising ongoing challenges for accurately assessing future model behavior (techcrunch.com).