Anthropic's newest and most powerful publicly available model just set a record nobody was rooting for. Artificial Analysis, the independent AI benchmarking firm, ran Claude Fable 5 through its Intelligence Index and came out with a $6,228 invoice , the most expensive model it has ever benchmarked, by a wide margin. That's 1.7x the cost of running Claude Opus 4.8 ($3,700) and 2.2x the cost of GPT-5.5 ($2,900). The number is a stark illustration of a new pricing reality Anthropic has quietly introduced with its Mythos-class tier.
A new tier above Opus
Claude Fable 5 is a Mythos-class model that Anthropic has made safe for general use, and its capabilities exceed those of any model the company has ever made generally available. Mythos-class models are a tier of Claude models that sit above the Opus class in capability. The first, Claude Mythos Preview, was released in April through Project Glasswing. Fable 5 and Mythos 5 followed.
The two models , Fable 5 and Mythos 5 , are not separate architectures. Anthropic states that they share the same underlying model and the same published specifications. The key difference is guardrails: in high-risk areas like cybersecurity, biology, chemistry, and distillation, Fable 5 blocks responses and falls back to Claude Opus 4.8. Mythos 5 has those restrictions lifted and is only available to a narrow group of government-adjacent partners through Project Glasswing.
The price tag, unpacked
The $6,228 benchmark cost isn't arbitrary. Artificial Analysis's v4.1 methodology accounts for real-world caching behavior, which is critical for agentic workloads. Here's how the number breaks down:
- 676M total tokens consumed: 590M input, 76.5M reasoning, 9.8M answer tokens
- Without caching: the bill would have been ~$10,215 at list price
- Cache savings: most input tokens were cached, billed at $1/M (reads) rather than $10/M, cutting input cost from ~$5,900 to ~$2,050
- Fallback credit: ~8% of tasks were served by Claude Opus 4.8 at $5/$25, trimming ~$275 off the total
Don't miss what's next in AI
Join 300,000+ engineers and researchers who get the signal, not the noise.
- Full access to in-depth AI research breakdowns
- Be the first to know what's trending before it hits mainstream
- Daily curated papers, repos, and industry moves