On June 9, 2026, Anthropic dropped Claude Fable 5 into the wild β€” its first publicly available model in a new Mythos class that sits above Opus in raw capability. The same underlying system also shipped as Claude Mythos 5, but with safeguards stripped away for a select group of cyberdefenders and infrastructure providers operating under Project Glasswing, a collaboration with the U.S. government. This isn't an incremental update to Opus 4.8; it's a deliberate ladder rung above anything Anthropic has previously released broadly.

The Mythos Class: What's Actually New Here

Anthropic's naming scheme carries weight if you dig into it. "Fable" derives from Latin fabula β€” that which is told β€” while its Greek counterpart mythos shares etymological DNA with the sibling model's name. This isn't marketing fluff; it's a signal that Mythos-class models represent a structural step forward, not a patch cycle on existing architecture. Claude Mythos Preview already shipped quietly in April through Project Glasswing, meaning this week's public release is the expansion of a program that began months ago under controlled conditions.

Benchmarks Don't Tell the Whole Story

Anthropic claims Fable 5 achieves state-of-the-art results across nearly every tested benchmark β€” software engineering, knowledge work, vision, and scientific research all show gains. CNBC reports it scored more than 10% higher than Opus 4.8 on some benchmarks, which is substantial for a single generation jump. But the number that actually matters isn't a static score; it's how performance degrades as tasks get longer. Anthropic emphasizes that Fable 5's advantage compounds with complexity β€” multi-step agent workflows, large codebases, extended research sessions. That's where this model changes what's possible versus its predecessors.

The Real Story: A Safety Split

Here's what makes this launch worth your attention. Fable 5 and Mythos 5 are the same underlying system. Same weights. Same architecture. What separates them is a conservative set of guardrails that Anthropic ships with the public version β€” in cybersecurity, biology, chemistry, and chemical distillation, Fable 5 declines requests and falls back to Opus 4.8 responses instead. These safeguards trigger in under 5% of sessions on average, which sounds low until you realize they're tuned conservatively enough to occasionally catch harmless queries too.

No Universal Jailbreaks Found β€” This Time

Before release, Anthropic ran an internal bug bounty targeting their classifier systems. Over more than 1,000 hours of red-teaming produced no universal jailbreak β€” meaning no technique that reliably bypassed safety measures across the board. External security organizations also came up empty-handed. That doesn't mean the model is unjailbreakable; it means Anthropic has raised the bar significantly higher than previous iterations. Mythos 5, with those safeguards lifted in restricted domains, went only to vetted cyberdefenders and infrastructure providers β€” initially through Project Glasswing, which increasingly looks like Anthropic's preferred channel for sensitive deployments.

Pricing and Access: Premium Tier Comes Premium Price

Fable 5 costs $10 per million input tokens and $50 per million output tokens. That's roughly double Opus 4.8 pricing but less than half what Mythos Preview commanded at launch β€” suggesting Anthropic is learning how to price its premium tiers more aggressively as demand outstrips supply. The company expects Fable 5 access to be "very high, and difficult to predict," which explains why subscription plans will roll out in stages rather than all-at-once availability.

Timing Is Never Incidental

Fable 5 dropped just days after Anthropic confidentially filed its IPO prospectus with the SEC. The company's revenue run rate now sits at roughly $47 billion, up from about $10 billion a year earlier β€” growth that makes even OpenAI's numbers look modest by comparison. Anthropic recently closed a funding round valuing the company at $965 billion, surpassing OpenAI's March valuation of $852 billion. Releasing a new premium tier with an explicit safety narrative into this moment is calculated positioning: it demonstrates both capability leadership and responsible deployment practices ahead of public market scrutiny.

Key Takeaways

  • Claude Fable 5 represents the first Mythos-class model publicly available, sitting above Opus in Anthropic's hierarchy
  • The same underlying system powers restricted Claude Mythos 5 for vetted government partners under Project Glasswing
  • Benchmarks show >10% gains over Opus 4.8 on some tests, with advantages that grow on longer tasks
  • Conservative safety guardrails trigger in <5% of sessions but will occasionally catch benign requests
  • Internal bug bounty found no universal jailbreaks across 1,000+ hours of testing before launch
  • Pricing at $10/$50 per million tokens is double Opus 4.8 but less than half Mythos Preview's cost

The Bottom Line

Anthropic just made capability and access separate products β€” same model, different safety profiles, different price tiers for different audiences. That's not incremental progress; it's a structural shift in how frontier AI gets distributed. If IPO day brings the kind of scrutiny this launch suggests Anthropic is preparing for, expect more tiered deployments where raw capability lives behind access controls calibrated to perceived risk.