Official source: anthropic.com
CLAUDE FABLE 5
Anthropic just shipped the most capable model you can actually use today — the first Mythos-class model released to the public. Here is what it is, what is new, and what the 2× price tag really buys you.
- What it is: Claude Fable 5 is the first public Mythos-class model — a tier that sits above the Opus class in raw capability.
- Why it matters: It is state of the art on nearly every benchmark, and its lead grows the longer and more complex the task.
- Price: $10 / million input tokens and $50 / million output tokens — officially confirmed at exactly 2× Opus 4.8 ($5 / $25), and the same price as Mythos 5.
- The catch: High-risk queries (cybersecurity, biology & chemistry, distillation) silently fall back to Opus 4.8. This triggers in under 5% of sessions.
- Access: Free on Pro, Max, Team & Enterprise seats through June 22; usage credits required from June 23. Fully live on the API today.
What exactly is Claude Fable 5?
Fable 5 is the first model from Anthropic's new Mythos class to reach the general public. Until now, Mythos-class capability lived behind closed doors — the original Mythos Preview launched in April through Project Glasswing, restricted to a handful of cyber-defence and critical-infrastructure partners.
Mythos is a tier that sits above the Opus class. So in plain terms: Fable 5 is more capable than Opus 4.8, which until today was the most powerful model you could touch. Anthropic describes its capabilities as exceeding any model they have ever made generally available, with state-of-the-art results across software engineering, knowledge work, vision, and scientific research.
The pattern to remember: the longer and harder the task, the bigger Fable 5's lead. On short, simple prompts the gap to Opus is modest. On long-horizon, multi-step, agentic work, it pulls away.
The new capabilities, division by division
Software engineering
In early testing, an enterprise partner had Fable 5 perform a codebase-wide migration on a 50-million-line codebase in a single day — work that would have taken a full team over two months by hand. It is also more token-efficient than past models.
Knowledge work
Top scores on senior-level finance reasoning benchmarks, with big gains in document reasoning, chart and table interpretation, and root-cause analysis. One analytics partner reported the first model to break 90% on its core long-running analytics benchmark — a 10-point jump over Opus.
Vision
The new state of the art for vision tasks. It can read precise numbers off detailed scientific figures and rebuild a web app's source code from screenshots alone — and it needs far less scaffolding to do it.
Memory & long context
Stays coherent across millions of tokens on long-running tasks and improves its own output using file-based notes. Persistent memory boosted its game-playing performance roughly three times more than it did for Opus 4.8.
Pricing: the 2× question, answered
This is the line that matters for anyone running real workloads. Per the official Claude pricing page, Fable 5 costs exactly double Opus 4.8 on both input and output — yet still less than half the price of the old Mythos Preview. Here is the full current lineup:
Source: claude.com/pricing. Prompt-cache writes cost $12.50 / MTok on Fable 5 vs $6.25 on Opus 4.8. Batch processing cuts any of these rates by 50%. When the safeguards route a request to Opus 4.8, those tokens are billed at Opus rates — the 2× premium only applies to work Fable does itself.
Fable 5 vs Opus 4.8 cost calculator
Estimate your monthly bill. Slide your token volume and see the premium in real dollars.
Rates: Fable 5 $10 in / $50 out per MTok; Opus 4.8 $5 / $25. Estimates exclude prompt caching, which can cut input costs dramatically on repeated context.
The safeguards: when Fable hands off to Opus
Fable 5 is not "Mythos with the lid off." Because a model this capable carries real misuse risk, Anthropic shipped it with classifiers that detect high-risk queries and quietly route them to Opus 4.8 instead. You get notified when it happens. The safeguards are deliberately tuned to be cautious — meaning they will occasionally catch a harmless request.
Cybersecurity
Vulnerability discovery, exploitation, and agentic offensive tasks. The classifiers block Fable from making progress on these and defer to Opus 4.8.
Biology & chemistry
Broadly routed to Opus 4.8 for now — a conservative choice Anthropic plans to narrow as it builds a trusted-access program for legitimate researchers.
Distillation
Requests flagged as attempts to "distill" Fable's abilities to train competing models also fall back to Opus 4.8.
For most people this is a non-issue: more than 95% of Fable sessions involve no fallback at all. For those sessions, you are effectively getting the full Mythos 5 experience.
Fable 5 and Mythos 5 are the same brain
The only difference is the guardrails. Fabula ("that which is told") is the Latin cousin of the Greek mythos — the naming is the tell. Same model, two doors.
Fable 5
- Available to everyone today
- Safeguards active → high-risk queries route to Opus 4.8
- API, Pro, Max, Team, Enterprise
Mythos 5
- Restricted to vetted partners
- Cyber safeguards lifted (Glasswing); biology access coming for select researchers
- Strongest cybersecurity capability of any model in the world
Availability & the subscription clock
The API and consumption-based Enterprise plans have Fable 5 live and fully available right now (model string claude-fable-5). Subscriptions are rolling out more conservatively because demand is expected to be very high:
Should you switch?
If your work is long-horizon and complex — large codebase migrations, multi-step agentic pipelines, dense analytical runs — Fable 5 is worth the premium, because that is exactly where its lead is largest and the 2× cost is most easily justified by fewer turns and faster completion.
For routine prompts, short tasks, and high-volume automated workflows, Opus 4.8 at $5 / $25 remains the value pick — and Sonnet 4.6 at $3 / $15 covers the lightest tier. The smart play is to route by difficulty: send your hardest jobs to Fable 5 and let Opus 4.8 carry the everyday load — especially since output tokens at $50/M are what add up fastest.
Claude Fable 5 — frequently asked questions
Is Claude Fable 5 the most capable Claude model available?
Yes. Fable 5 is the most capable model Anthropic has ever made generally available. It belongs to the Mythos class, a capability tier above Opus, and is state of the art on nearly every tested benchmark — with the biggest lead on long, complex, agentic tasks.
How much does Claude Fable 5 cost?
On the API, $10 per million input tokens and $50 per million output tokens — exactly twice the price of Claude Opus 4.8 ($5 / $25). Prompt-cache reads cost $1 / MTok, cache writes $12.50 / MTok, and batch processing cuts all rates by 50%.
What is the difference between Fable 5 and Mythos 5?
They are the same underlying model at the same price. Fable 5 ships with safeguards that route high-risk queries (cybersecurity, biology and chemistry, distillation) to Opus 4.8 and is available to everyone. Mythos 5 has those safeguards lifted in some areas and is restricted to vetted partners, mainly through Project Glasswing.
Will my requests actually get downgraded to Opus 4.8?
Rarely. Anthropic's early data shows more than 95% of Fable sessions involve no fallback at all. When a fallback does happen you are notified, and those tokens are billed at the cheaper Opus rate, not the Fable premium.
Is Fable 5 included in Claude Pro and Max subscriptions?
Temporarily, yes. From June 9 through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans as promotional access. From June 23 it requires usage credits, and Anthropic aims to restore it as a standard plan feature once capacity allows.
When should I use Fable 5 instead of Opus 4.8?
Use Fable 5 for your hardest, longest jobs: large codebase migrations, multi-step agentic pipelines, dense document analysis, and complex research. For routine prompts and high-volume automation, Opus 4.8 (or Sonnet 4.6 at $3 / $15) delivers better value. Route by difficulty, not by default.
Want the primary source?
Read Anthropic's official announcement