Claude Fable 5 is the rare model launch that made headlines twice — once for how capable it is, and once for briefly disappearing. Anthropic released it on June 9, 2026, pulled it three days later to comply with U.S. export controls, and brought it back everywhere on July 1. Now that it’s live and here to stay, it’s worth understanding what all the noise was about.
Short version: Fable 5 is the most powerful model Anthropic has ever made generally available. It sits in a brand-new tier — “Mythos-class” — that ranks above the Opus line that used to top the Claude family. For anyone building with AI, that reshuffles the deck. Here’s what Fable 5 actually does, how it compares, what it costs, and when it’s the right tool to reach for.
What Is Claude Fable 5?
Claude Fable 5 is a Mythos-class model tuned for long, complex, autonomous work — the kind of multi-day tasks that previous models couldn’t sustain without losing the plot. Think large codebase migrations, deep research, multi-stage analysis, and agent workflows that run for hours or days with minimal supervision.
The headline trait is stamina. Fable 5 stays coherent across millions of tokens, takes notes to improve its own output, and tests its own work as it goes. Anthropic’s framing is blunt: the longer and more complex the task, the bigger Fable 5’s lead over everything else.
Fable 5 vs. Mythos 5: Why Two Models?
Fable 5 and Mythos 5 are the same underlying model. The only difference is safeguards — and that difference is the whole point of the two names (“Fable,” from the Latin fabula, is a cousin of the Greek mythos).
- Claude Fable 5 is the public release. It ships with strong safeguards for high-risk domains and is available to Pro, Max, Team, and Enterprise users, plus developers on the API.
- Claude Mythos 5 is the restricted version, with some safeguards lifted, available only to vetted cyber defenders and infrastructure providers through Anthropic’s Project Glasswing — plus, soon, a small set of biology researchers.
In plain terms: Fable is the version you can use today; Mythos is the less-restricted twin kept behind a trusted-access wall.
How Fable 5 Performs
Across nearly every published benchmark, Fable 5 leads — beating Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro on agentic coding, knowledge work, vision, tool use, and reasoning. It posts 80.3% on SWE-Bench Pro, 88.0% on Terminal-Bench 2.1, and a category-topping 1932 on the GDPval-AA knowledge-work benchmark.
A note on the asterisks: on cybersecurity and biology benchmarks (the starred rows), Fable 5’s published scores are lower because its safeguards deliberately route those queries elsewhere. That’s not a capability gap — it’s a safety design choice, which brings us to the most important thing to understand before you use it.
The Safeguards (and the Opus 4.8 Fallback)
Fable 5 is powerful enough that Anthropic gated it. When a request touches cybersecurity, biology, chemistry, or model-distillation attempts, a separate classifier intercepts it and hands the response to Claude Opus 4.8 instead. You’ll be told when this happens, and you won’t pay Fable prices for a rerouted request.
Two practical things to know. First, the safeguards are tuned conservatively, so they occasionally catch harmless questions — Anthropic says fallback triggers in under 5% of sessions on average, meaning 95%+ of sessions run on full Fable. Second, a fallback to Opus 4.8 isn’t a refusal; Opus is a strong model in its own right, so you still get a real answer. If your work lives in security or life sciences, factor this routing in.
There’s also a data policy worth flagging: using Fable requires 30-day data retention for safety monitoring, on both first- and third-party surfaces.
What Fable 5 Is Actually Good At
Benchmarks are one thing; real workloads are another. The early signal across Anthropic’s testing and partners points to four standout areas:
- Coding. Stripe used Fable 5 to run a migration across a 50-million-line Ruby codebase in a single day — work estimated at more than two months by hand. It also tops Cognition’s FrontierCode eval and is more token-efficient than past Claude models.
- Knowledge work. It leads on Hebbia’s senior-level finance benchmark and, per CNBC, cleared Opus 4.8 by more than 10 points on some tests — with real gains in chart, table, and document reasoning.
- Vision. It can pull exact numbers from scientific figures and rebuild a web app’s source code from a screenshot alone — and it beat Pokémon FireRed using vision only, no helper scaffolding.
- Memory. Given file-based notes in Slay the Spire, it improved three times more than Opus 4.8 and reached the final act three times as often — a proxy for how well it sustains long, stateful tasks.
What It Costs and Who Can Get It
Fable 5 runs $10 per million input tokens and $50 per million output tokens — roughly 2x Opus 4.8, but under half the price of the earlier Mythos Preview. It’s available on Pro, Max, Team, and seat-based Enterprise plans, plus the Claude API (claude-fable-5) and the major clouds.
The pricing logic is “cheaper per finished task, even if pricier per token.” That can hold for hard engineering and research work, where a stronger model wastes less time — but long agent sessions burn a lot of tokens, so watch your usage on open-ended runs.
Should You Use It?
Reach for Fable 5 when the job is genuinely hard and long: big migrations, complex implementations, multi-day agent runs, or deep analysis where a weaker model costs you more in wasted cycles than Fable costs in tokens. For quick drafting, summarization, and everyday assistance, a cheaper, faster model is the smarter default. Match the model to the difficulty of the task, not the other way around.
Key Takeaways
- Fable 5 is Anthropic’s most capable public model — a new “Mythos-class” tier that sits above Opus 4.8.
- It’s built for long, autonomous work and leads nearly every benchmark against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.
- Safeguards route cyber/bio/chem queries to Opus 4.8 in under 5% of sessions; you’re notified and not charged Fable rates for those.
- Pricing is $10/$50 per million tokens — premium, but justified when task difficulty is high.
Ready to Pick the Right Model?
Trying to figure out which model fits your workflow — Fable 5, Opus 4.8, or something lighter? Explore our up-to-date model comparisons and tool breakdowns at aitoolcurator.com to pick the right AI for the job.
Claude Fable 5 FAQ
Is Claude Fable 5 available now?
Yes. After a brief suspension in June 2026 to meet U.S. export-control rules, Anthropic restored full access on July 1, 2026. Fable 5 is now available to Pro, Max, Team, and Enterprise users, plus developers via the Claude API and major cloud platforms.
What’s the difference between Claude Fable 5 and Mythos 5?
They’re the same underlying model — only the safeguards differ. Fable 5 is the public release with full safety protections, while Mythos 5 has some safeguards lifted and is restricted to vetted cyber defenders and infrastructure providers through Anthropic’s Project Glasswing.
How much does Claude Fable 5 cost?
Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens — roughly twice the price of Opus 4.8, but under half what the earlier Mythos Preview cost. It’s positioned as cheaper per finished task, even if pricier per token.
Is Claude Fable 5 better than Opus 4.8, GPT-5.5, and Gemini 3.1 Pro?
On published benchmarks, yes. Fable 5 leads nearly every category — agentic coding, knowledge work, vision, tool use, and reasoning — including 80.3% on SWE-Bench Pro and 88.0% on Terminal-Bench 2.1. For short, simple tasks, though, a cheaper model is still the smarter default.
Sources: Anthropic — Claude Fable 5 and Mythos 5 announcement · Anthropic — Claude Fable product page
Model details, pricing, and availability accurate as of July 5, 2026, and subject to change.
