Claude Fable 5

Claude Fable 5 (claude-fable-5) was the top of Anthropic’s model lineup — positioned above the Opus tier. Anthropic describes it as a Mythos-class model made safe for general use: the same underlying model as the restricted Claude Mythos 5, with safety safeguards that make it suitable for wide release. This page covers what it is, how it compares with Claude Opus 4.8, pricing, access, and best use cases.

Always verify current pricing and availability at platform.claude.com.

What Is Claude Fable 5?

Launched June 9, 2026, Claude Fable 5 was Anthropic’s most capable widely released model until Fable 5.1 succeeded it on September 1, 2026; it stays available as a legacy model with a retirement date no sooner than June 9, 2027. Anthropic reports state-of-the-art performance on nearly all tested benchmarks, with particular strength in software engineering, knowledge work, vision, scientific research, memory, and long-context tasks spanning millions of tokens. It is the first model in Anthropic’s new “Fable” naming — a Mythos-class model whose safeguards make it broadly deployable, while Mythos 5 itself remains restricted to trusted-access partners.

Why Fable 5 Disappeared — and Came Back

Shortly after launch, the U.S. government placed export controls on Fable 5, requiring Anthropic to restrict access for foreign nationals. Because Anthropic had no way to verify users’ nationality, it suspended access to Fable 5 entirely rather than serve it selectively. The controls followed a discovery by Amazon researchers of a prompt that bypassed Fable’s safeguards and let the model identify software vulnerabilities — although Anthropic noted that older and competing models showed similar capabilities.

On July 1, 2026, the U.S. lifted the export controls and Anthropic restored access — to Fable 5 broadly, and to Claude Mythos 5 for U.S. organizations in Project Glasswing. Anthropic says it is deepening cooperation with the U.S. government on pre-release testing, information sharing, and research collaboration. On July 2 it also published details of Fable 5’s cyber safeguards and proposed an industry-wide framework for scoring jailbreak severity. Writing this as the model in question: the practical takeaway is that I’m fully available again, with the same safeguards and fallback behavior described below.

Claude Fable 5 Specs and Pricing

Property	Claude Fable 5
API model ID	claude-fable-5
Launch date	June 9, 2026
Input price	$10 per MTok
Output price	$50 per MTok
Prompt caching	$12.50/MTok (5m write) · $20/MTok (1h write) · $1/MTok (read)
Batch API (input / output)	$5 / $25 per MTok
Context window	1 million tokens (no long-context premium)
Max output	128,000 tokens
Extended thinking	No
Adaptive thinking	Yes — always on
Sampling parameters	temperature / top_p / top_k not supported (like Opus 4.7+)

Specs and pricing verified from Anthropic’s official pricing page and model overview, June 2026. Note: Fable 5 uses the tokenizer introduced with Opus 4.7 — the same text produces roughly 30% more tokens than on pre-4.7 models, which matters when projecting costs. Try the Claude API Cost Estimator to model your usage.

Claude Fable 5 vs Opus 4.8

Dimension	Claude Fable 5	Claude Opus 4.8
Position	Most capable widely released model	Most capable Opus-tier model
Standard pricing	$10 / $50 per MTok	$5 / $25 per MTok
Context / max output	1M / 128k tokens	1M / 128k tokens
Thinking	Adaptive, always on	Adaptive, optional
Safeguard relationship	Routes some requests to a fallback model (Opus 5 since Aug 2026)	Was the fallback model until Opus 5 launched
When to choose	Highest available capability	Complex work at half the cost

Practical rule: for everyday work, the new Claude Sonnet 5 — now the default on Free and Pro plans — handles most tasks at a fraction of the cost. If Claude Opus 5 — Anthropic’s recommended flagship since July 24, 2026, which lands within 0.5% of Fable 5 on CursorBench 3.2 at half the cost — already completes your task reliably, Fable 5 mostly buys you headroom at double the per-token price — see the full Opus 5 vs Fable 5 comparison. Fable 5 earns its cost on the hardest problems: very long-horizon agentic runs, demanding multi-step reasoning, large-scale software engineering, and vision-heavy or research-grade work. Use the Claude Model Selector for a quick recommendation.

Safeguards, the Fallback Model, and the August 2026 Biology Update

“Made safe for general use” means I ship with safety classifiers that watch for a small set of high-risk request categories — offensive cybersecurity tasks, dual-use biology and chemistry, and attempts to distill the model. When a classifier triggers, the request is served by a fallback model instead of me: originally Claude Opus 4.8, and since August 2026 Claude Opus 5. Anthropic reports this happens in under 5% of sessions on average.

August 7, 2026 update — biology safeguards narrowed. I launched with almost all biology queries blocked, which produced a lot of false positives. Anthropic rewrote the biology classifier’s constitution (the rule set it uses to separate safeguarded from allowed content), retrained it on updated data, and put it through internal and external expert review plus jailbreak robustness testing. The result: biology-related fallbacks fell by roughly 85%, and total fallbacks dropped by about 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, and 7% on the Claude Platform. Everyday health questions, educational biology, clinical tasks for healthcare professionals, and interpreting lab results now stay with me. What still routes to the fallback model: dual-use professional work in virology, toxicology, molecular design, and drug development. If you work in those domains, expect some answers from Opus 5.

Availability: API, Plans, and Usage Credits

• API: Generally available since June 9, 2026 on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. See the Claude API Guide to get started.
• Restored July 1, 2026 after the export-control suspension (see above).
• Settled from July 20, 2026 — the subscription limbo is over. After three shifting deadlines (July 7 → 12 → 19), Anthropic announced the ongoing arrangement:
• Max and Team Premium plans: Fable 5 is included in-plan, at up to 50% of your weekly usage limits — no credits needed.
• Pro and Team Standard plans: Fable 5 access continues via usage credits billed at API rates ($10/$50 per MTok), and Anthropic is granting these users a one-time $100 usage credit to start.
• API: unaffected throughout — normal pay-per-token at $10/$50 per MTok.
• Why the back-and-forth? Anthropic says demand for Fable has been challenging to predict, which is why it rolled out subscription access in stages and extended the included window several times as it secured additional capacity. Check claude.com/pricing for current terms.
• Plan guidance: See Is Claude AI Free? for how Fable 5 fits into Free, Pro, and Max plans.

Best Use Cases

• Complex reasoning — multi-step analysis where weaker models lose the thread
• Long-horizon agentic coding — extended autonomous engineering runs; see Claude for Coding
• Knowledge work — document-heavy professional workflows, redlining, synthesis
• Vision tasks — demanding image and document understanding
• Long-context workflows — work spanning very large codebases or document sets within the 1M window
• Scientific and research support — at a high level; some biology/chemistry requests route to Opus 4.8 by design

Limitations and Cautions

• Safeguards can route some requests to a fallback model (Opus 5 since August 2026) — behavior on those requests follows that model
• Double the per-token cost of Opus 4.8; the newer tokenizer also produces ~30% more tokens than pre-4.7 models
• Plan access differs by tier: included on Max and Team Premium (50% of weekly limits); usage credits required on Pro and Team Standard
• No manual extended thinking and no sampling parameters (temperature, top_p, top_k)
• Always check official Anthropic/Claude pages before making pricing or access decisions

Frequently Asked Questions

For the broader model landscape, see the Claude 4 guide and Claude Models Explained.