Claude Fable 5: Anthropic's New Flagship AI Model Explained (Pricing, Benchmarks & Access)

Anthropic just unveiled Claude Fable 5, its most capable model to date — built for the kind of long, complex, autonomous work that previous Claude models couldn’t sustain. Alongside it comes Claude Mythos 5, the same model with its safety guardrails removed for a tightly controlled set of security and research partners.

Announced on June 9, 2026, Fable 5 introduces a new top tier in the Claude family — a “Mythos-class” that sits above the Opus line. Here’s what it is, what it costs, how it benchmarks, and how it fits into Anthropic’s 2026 model lineup.

What is Claude Fable 5?

Anthropic describes Fable 5 as “the next generation of intelligence for the hardest knowledge work and coding problems.” The headline capability is endurance: Fable 5 is built to handle days-long, complex, asynchronous tasks that earlier models couldn’t sustain, working autonomously for longer than any previous Claude.

Three things stand out:

• Long-horizon autonomy — it runs for extended stretches inside agent frameworks like Claude Code without losing the thread, which Anthropic frames as its biggest leap: “the longer and more complex the task, the larger Fable 5’s lead.”
• Vision — strong understanding of diagrams, charts, and tables inside documents, useful for knowledge work where the data lives in a screenshot or a PDF.
• Self-validation — Fable 5 tests and checks its own work rather than handing back a first draft.

Fable 5 vs Mythos 5: same brain, different guardrails

This is the part that makes the launch unusual. Fable 5 and Mythos 5 are the same underlying model. The difference is entirely about safety:

• Claude Fable 5 ships with safeguards active in three areas — cybersecurity, biology/chemistry, and model distillation. It’s the version anyone can use.
• Claude Mythos 5 is that same model with the safeguards lifted in those areas. It is not public: access runs through Project Glasswing, a US-government collaboration that puts the unrestricted model in the hands of cyberdefenders and critical-infrastructure providers, plus a small number of vetted biology researchers.

Anthropic says Project Glasswing — which began with the earlier Claude Mythos Preview — is being expanded to roughly 150 new organizations across more than fifteen countries.

How the safety routing works

Rather than refuse sensitive prompts outright, Fable 5 reroutes them. When the model detects:

• offensive cyber, exploitation, or agentic hacking requests,
• most biology and chemistry requests, or
• attempts to distill the model,

…it falls back to Claude Opus 4.8 to answer. Crucially, those routed requests are not billed at Fable pricing. Anthropic reports that safeguards trigger in less than 5% of sessions — so for ordinary coding and knowledge work, you’re talking to Fable 5 the whole way.

The benchmarks: where Fable 5 wins

Anthropic leans on long-horizon and domain-specific evaluations rather than the usual leaderboard one-liners:

• FrontierCode — highest among frontier models, even at medium effort.
• Hebbia Finance Benchmark — the highest score of any model tested.
• IMC trading-analysis evaluations — “aced their trading-analysis evaluations nearly across the board.”
• Core analytics benchmark — the first model to break 90%, a 10-point jump over Opus.

The customer proof points are more striking than the numbers:

Stripe reported that Fable 5 “compressed months of engineering into days,” migrating a 50-million-line Ruby codebase in a single day — work that had taken two months by hand.

Cursor CEO Michael Truell said Fable 5 “opened up a class of long-horizon problems that were out of reach for earlier models.”

GitHub CPO Mario Rodriguez reported it exceeded previous benchmarks on complex, long-horizon coding tasks.

Pricing: a premium tier above Opus

Fable 5 is priced as a flagship:

• $10 per million input tokens
• $50 per million output tokens
• Prompt caching cuts repeated-context cost by up to 90%
• A US-only inference option is available at a 1.1x multiplier

Anthropic frames this as less than half the price of the earlier Claude Mythos Preview — the premium tier is getting cheaper as it goes mainstream. It’s still double the price of Opus 4.8 ($5 / $25), which is the trade-off for the extra capability and endurance.

Where Fable 5 sits in the Claude lineup

Fable 5 doesn’t replace the existing models — it tops them. Here’s the full 2026 stack:

Model	Tier	Input / Output (per 1M)	Context	Best for
Claude Fable 5	Mythos-class (new)	$10 / $50	Extended	The hardest, longest-horizon coding & knowledge work
Claude Opus 4.8	Opus	$5 / $25	1M	Complex agentic & coding work; Fable’s safety fallback
Claude Sonnet 4.6	Sonnet	$3 / $15	1M	Best balance of speed and intelligence
Claude Haiku 4.5	Haiku	$1 / $5	200K	Fastest, most cost-effective for simple tasks

The mental model: Haiku for speed and volume, Sonnet for everyday balance, Opus 4.8 for hard problems, and Fable 5 when the task is so long or so complex that nothing else holds up — with Opus 4.8 quietly backstopping it on anything sensitive.

How to access Claude Fable 5

• API — fully available from launch on the Claude Platform.
• Cloud — available on Amazon AWS, Google Cloud, and Microsoft Foundry.
• Subscriptions — for Pro, Max, Team, and Enterprise plans, Fable 5 is included from June 9 to June 22, 2026. From June 23, usage beyond your plan limits switches to usage credits.

What this launch means

Two signals matter here. First, the frontier is now measured in days, not turns — Anthropic’s whole pitch for Fable 5 is sustained autonomous work, and the Stripe migration is the kind of proof point that reframes what “an AI coding tool” can do.

Second, the Fable / Mythos split is a new safety pattern: ship the capable model to everyone, keep the unrestricted twin behind government-grade access controls, and silently route the dangerous 5% of prompts down to a more conservative model instead of refusing them. Whether that holds up is the story to watch — but as a product design, it’s a notable departure from a simple “the model says no” guardrail.

Key takeaways

• Launch date: June 9, 2026
• What it is: Anthropic’s most capable model, a new “Mythos-class” tier above Opus
• Pricing: $10 / $50 per million tokens, up to 90% off with prompt caching
• Fable 5 vs Mythos 5: same model; Mythos 5 has safeguards removed and is restricted to Project Glasswing partners
• Safety routing: sensitive cyber/bio/distillation requests fall back to Opus 4.8, free of Fable pricing, in under 5% of sessions
• Access: API + AWS + Google Cloud + Microsoft Foundry today; subscriptions included through June 22

---

Sources

• Anthropic — Claude Fable 5 overview
• Anthropic — Introducing Claude Fable 5 and Claude Mythos 5

Frequently asked questions

What is Claude Fable 5? Claude Fable 5 is Anthropic’s newest frontier model, announced on June 9, 2026. Anthropic positions it as the next generation of intelligence for the hardest knowledge work and coding problems, with the ability to sustain days-long, complex, asynchronous tasks that previous models couldn’t. It belongs to a new “Mythos-class” tier that sits above the Opus class in capability, and it ships with strong vision and self-validation abilities.

How much does Claude Fable 5 cost? Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, with prompt caching cutting the cost of repeated context by up to 90%. Anthropic notes this is less than half the price of the earlier Claude Mythos Preview. For comparison, Claude Opus 4.8 is $5 input / $25 output per million tokens.

What is the difference between Claude Fable 5 and Claude Mythos 5? Fable 5 and Mythos 5 are the same underlying model. Fable 5 is the public version with safety safeguards active in cybersecurity, biology/chemistry, and model distillation. Mythos 5 is the same model with those safeguards lifted, restricted to vetted cyberdefenders and select researchers through Anthropic’s Project Glasswing and a trusted-access program.

How does Claude Fable 5 compare to Claude Opus 4.8? Fable 5 sits one tier above Opus 4.8 and is used for the hardest, longest-horizon tasks. Anthropic uses Opus 4.8 as the benchmarking baseline and as the automatic fallback model: when Fable 5’s safeguards flag a sensitive request (offensive cyber, most biology/chemistry, or distillation attempts), the query is routed to Opus 4.8 and is not charged at Fable pricing. Anthropic says safeguards trigger in less than 5% of sessions.

How do I access Claude Fable 5? Claude Fable 5 is available immediately through the Claude API, and on Amazon AWS, Google Cloud, and Microsoft Foundry. For Pro, Max, Team, and Enterprise subscribers, Fable 5 usage is included from June 9 to June 22, 2026; from June 23 onward, usage beyond plan limits requires usage credits. Mythos 5 remains restricted to Project Glasswing partners and select biology researchers.

What are Claude Fable 5’s benchmark results? According to Anthropic, Fable 5 scores highest among frontier models on the FrontierCode evaluation at medium effort, posts the highest score of any model tested on the Hebbia Finance Benchmark, aced IMC’s trading-analysis evaluations nearly across the board, and is the first model to break 90% on a core analytics benchmark — a 10-point jump over Opus. Stripe reported it migrated a 50-million-line Ruby codebase in a single day, work that took two months manually.

What are the other Claude models in 2026? Below Fable 5, Anthropic’s lineup is Claude Opus 4.8 ($5/$25 per million tokens, 1M context) for complex agentic and coding work, Claude Sonnet 4.6 ($3/$15, 1M context) for the best balance of speed and intelligence, and Claude Haiku 4.5 ($1/$5, 200K context) as the fastest and most cost-effective option for simple tasks.

Is Claude better than ChatGPT? It depends on the task, but Claude has a strong reputation for coding, long-document analysis, and natural writing, while ChatGPT (OpenAI) leads on some multimodal and ecosystem features. With Claude Fable 5 and Opus 4.8, Anthropic is widely seen as ahead on long-horizon, agentic coding — the kind of multi-step work tools like Claude Code rely on. For most developers the honest answer is to test both on your own workflow; for sustained autonomous coding, Claude’s 2026 lineup is hard to beat.

Which Claude model is best for coding? For the hardest, longest-running coding tasks, Claude Fable 5 is now the top choice. For most complex coding and agentic work, Claude Opus 4.8 is the workhorse. Claude Sonnet 4.6 is the best value for fast, everyday coding, and Claude Haiku 4.5 suits simple, high-volume tasks. All of them power Anthropic’s coding agent, Claude Code.

Is Claude AI free, and how much does it cost? Yes — claude.ai has a free tier. Paid consumer plans start with Claude Pro at around $20/month, with Max, Team, and Enterprise above it. Developers pay per token through the API, from Claude Haiku 4.5 ($1/$5 per million tokens) up to Claude Fable 5 ($10/$50). Fable 5 is included in Pro, Max, Team, and Enterprise plans from June 9 to June 22, 2026; after that, usage beyond plan limits is billed via usage credits.

Are Anthropic and Claude the same thing? No. Anthropic is the AI safety company, founded in 2021 by former OpenAI researchers including Dario Amodei. Claude is the family of AI models Anthropic builds — which in 2026 includes Claude Fable 5, Opus 4.8, Sonnet 4.6, and Haiku 4.5. In short: Anthropic is the company, Claude is the product.

Why is the AI called Claude? Claude is widely understood to be named after Claude Shannon, the American mathematician who founded information theory — the mathematical groundwork behind modern computing and AI. The name fits Anthropic’s research-first, safety-focused identity.

What’s the difference between Claude Opus, Sonnet, and Haiku — and which should I use? They’re three tiers of the same Claude family, trading cost for capability. Claude Haiku 4.5 ($1/$5 per million tokens, 200K context) is the fastest and cheapest — best for simple, high-volume tasks. Claude Sonnet 4.6 ($3/$15, 1M context) is the everyday balance of speed and intelligence. Claude Opus 4.8 ($5/$25, 1M context) is for the hardest reasoning and agentic coding. Above them sits the new Claude Fable 5 ($10/$50) for the longest, most complex work. Rule of thumb: Haiku for volume, Sonnet for daily work, Opus for hard problems, Fable 5 when nothing else holds up.

Can I access Claude Mythos 5? Not as a normal user. Claude Mythos 5 is the same model as Fable 5 but with safety guardrails removed, and access is restricted to vetted cyberdefenders and select researchers through Anthropic’s Project Glasswing and a trusted-access program — it’s not on the API or in consumer subscriptions. The model you can actually use is Claude Fable 5, which is publicly available with safeguards on.

Is Claude safe to use? For everyday use, Claude is regarded as one of the safer mainstream AI assistants — safety is core to Anthropic’s mission, and frontier models like Fable 5 route sensitive cyber and bio requests to a more conservative model instead of answering them. On privacy: Anthropic does not train on business and API data by default; for consumer plans, review the data-training toggle in your privacy settings. As with any AI, don’t paste secrets or regulated data.

Is Claude related to Elon Musk? No. Claude is built by Anthropic, an AI safety company founded in 2021 by former OpenAI researchers led by Dario Amodei. Elon Musk’s AI company is xAI, which makes Grok — a different product from a different company. The two are sometimes confused because both are major players in the 2026 AI race.

Why is Claude so expensive? Claude isn’t uniformly expensive — it spans cheap to premium. Claude Haiku 4.5 costs just $1/$5 per million tokens, while the new flagship Claude Fable 5 is $10/$50. The pricier tiers reflect the cost of running much larger, more capable models for hard reasoning and long, autonomous tasks. If cost matters, drop to Sonnet 4.6 or Haiku 4.5, or use prompt caching, which can cut repeated-context cost by up to 90%.