Claude Fable 5 Explained: Benchmarks, Pricing, and How to Use It

Claude Fable 5 explained — what the new Mythos-class model is, verified benchmarks vs Opus 4.8 and GPT-5.5, pricing, and how to start using it today.

• 1. What Is Claude Fable 5?
• 2. Claude Fable 5 vs Claude Mythos 5
• 3. Where Fable 5 Sits in the Claude Lineup
• 4. Claude Fable 5 Benchmarks
  • 4.1. Coding Benchmarks
  • 4.2. Knowledge Work Benchmarks
• 5. Pricing and Specifications
• 6. How to Use Claude Fable 5
  • 6.1. In the Claude App (claude.ai)
  • 6.2. In Claude Code
  • 6.3. From Your Own Application via the Claude API
• 7. What Changes for API Integrations
• 8. Should You Switch to Fable 5?
• 9. Conclusion
• Further Reading

 On June 9, 2026, Anthropic released Claude Fable 5, the most capable model it has ever made generally available. It is the first model in a new tier the company calls “Mythos-class” — a level above the familiar Opus, Sonnet, and Haiku families that millions of people already use through the Claude apps, Claude Code, and the Claude API.

Claude Fable 5 matters for a simple reason: the largest gains are in the areas where AI assistance has the most practical impact — agentic coding, long-running multi-step tasks, and complex analytical work. Anthropic states that the longer and more complex the task, the larger Fable 5’s lead over its other models.

In this article, we will look at what Claude Fable 5 actually is, how it relates to its restricted sibling Claude Mythos 5, the benchmark numbers published so far, pricing and specifications, and how you can start using it today from claude.ai, Claude Code, or your own application. A note on accuracy: the model is only days old, so every figure below comes from Anthropic’s launch materials, the official Claude API documentation, or attributed third-party reporting. Anthropic’s own documentation is always the authoritative source — if anything here is unclear or seems out of date, review the official pages linked at the end before acting on it.

1. What Is Claude Fable 5?

Claude Fable 5 is Anthropic’s newest frontier model and the first publicly available model in the Mythos-class tier. Anthropic describes it as state-of-the-art across nearly all tested AI capability benchmarks, with exceptional performance in software engineering, knowledge work, vision, and scientific research.

Compared with earlier Claude models, the defining characteristic is sustained autonomy. Fable 5 is built for long-running, ambiguous, multi-step tasks that previously required frequent human check-ins. The standout customer example comes from Anthropic’s launch announcement: during early testing, Stripe reported that in a 50-million-line codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.

The practical takeaway is that the kinds of work that used to need constant supervision — multi-file refactors, repository-wide changes, migration planning, deep document analysis — are precisely where this model class is strongest.

2. Claude Fable 5 vs Claude Mythos 5

Anthropic shipped two models at once, and the distinction is important to understand before you compare anything else.

• Claude Fable 5 is the generally available release. It includes safety classifiers covering three areas — cybersecurity, biology and chemistry, and model distillation. When a classifier flags a request in Claude’s products, the response is automatically handled by Claude Opus 4.8 instead, and the user is notified. Anthropic’s early data shows more than 95% of Fable sessions involve no fallback at all.
• Claude Mythos 5 is the same underlying model with certain safeguards lifted. It is not generally available: it is restricted to Project Glasswing partners (with cyber safeguards lifted) and, per Anthropic, soon to select biology researchers, until a broader trusted access program opens.

In other words, if you are a regular user or team, Claude Fable 5 is the model you can actually use. For everyday work, the safeguards rarely come into play, but API integrations should still plan for the possibility of a declined request, which we cover in Section 7.

3. Where Fable 5 Sits in the Claude Lineup

Here is how the current generally available Claude models compare at a glance. Fable 5 does not replace the other models; it sits above them as a new top tier.

Model	Tier	API Model ID	Best For
Claude Fable 5	Mythos-class (new)	claude-fable-5	Hardest reasoning, long-horizon agentic work, 1M-token context tasks
Claude Opus 4.8	Opus	claude-opus-4-8	Heavy coding and analysis at half the price of Fable 5
Claude Sonnet 4.6	Sonnet	claude-sonnet-4-6	Balanced everyday tasks
Claude Haiku 4.5	Haiku	claude-haiku-4-5	Fast, low-cost tasks like classification and summarization

The sensible workflow is escalation, not replacement. Route routine work to Sonnet or Haiku, heavy work to Opus 4.8, and reserve Fable 5 for the tasks where frontier reasoning genuinely changes the outcome — large migrations, difficult multi-step problems, and deep analytical work.

4. Claude Fable 5 Benchmarks

Anthropic published comparisons against Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro across roughly fifteen benchmarks. The numbers below are from Anthropic’s launch materials as reported by independent outlets including The Decoder and VentureBeat.

4.1. Coding Benchmarks

Benchmark	Fable 5	Opus 4.8	GPT-5.5	Gemini 3.1 Pro
SWE-Bench Pro (real GitHub engineering tasks)	80.3%	69.2%	58.6%	54.2%
Cognition FrontierCode Diamond (production-grade agentic coding)	29.3%	13.4%	5.7%	—

Two things stand out. First, an 11-point jump over Opus 4.8 on SWE-Bench Pro is a large generational gap — for comparison, the entire spread between Opus 4.8 and Gemini 3.1 Pro is 15 points. Second, the FrontierCode result more than doubles Opus 4.8’s score on a benchmark specifically designed around maintainable, production-quality code. Anthropic also reports that Fable 5 scores highest among frontier models on FrontierCode, even with medium reasoning effort, indicating stronger results without maximum token spend.

4.2. Knowledge Work Benchmarks

Benchmark	Fable 5	Opus 4.8	GPT-5.5	Gemini 3.1 Pro
GDPval-AA (enterprise knowledge work)	1932	1890	1769	1314
GDPpdf (visual document reasoning, no tools)	29.8%	22.5%	24.9%	16.7%

Early-access customer feedback on Anthropic’s launch page points the same direction: one partner reported that Fable 5 was the first model to break 90% on its core analytics benchmark of complex, long-running analytical tasks — a 10-point jump over Opus. Treat all of these early figures as launch-week numbers from the vendor and its partners; independent replication will take time, and you should always validate on your own workload before committing.

5. Pricing and Specifications

The official Claude API documentation lists the following specifications for Claude Fable 5.

Specification	Value
API model ID	claude-fable-5
Context window	1,000,000 tokens (standard pricing across the full window)
Max output tokens	128,000 per request
Input price	$10 per million tokens
Output price	$50 per million tokens
Thinking mode	Adaptive thinking only (disabling thinking is not supported)
Data retention	30 days; not available under zero data retention

That pricing is double Claude Opus 4.8, so cost discipline matters. Prompt caching is the key lever: cache hits are billed at a 90% discount on the input, making long-context agentic workflows far more affordable when the same large context is reused. Also remember that the 1M-token window is a capability, not a discount — a full-context request costs around $10 in input alone, so send the model what it needs, not everything it can read.

On the consumer side, Anthropic’s announcement confirms the rollout plan: Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22, 2026. From June 23, using it on those plans will require usage credits, though Anthropic says it will extend the included window if capacity allows and aims to restore Fable 5 as a standard part of subscription plans over time. On the Claude API and consumption-based Enterprise plans, it is fully available from day one.

6. How to Use Claude Fable 5

6.1. In the Claude App (claude.ai)

The simplest way to try Fable 5 is the model picker in the Claude web, desktop, or mobile apps. Select Claude Fable 5 from the dropdown and use it exactly as you would any other model. This is the right place to evaluate the quality of its reasoning on your own problems before changing anything in your tooling.

6.2. In Claude Code

If you work in a codebase, Claude Code is where Fable 5 is most interesting. Inside a Claude Code session, you can switch models with the /model command, or set the model when launching from your project root:

# Launch Claude Code with Fable 5 from your project root claude –model claude-fable-5   # Or switch inside an existing session /model claude-fable-5

The disciplined workflow stays the same: ask it to read existing files before writing, verify with your build and tests, and review every diff. The difference you should notice is in the long, complex tasks: large refactors and migrations that previously required many corrective prompts now converge in fewer steps.

6.3. From Your Own Application via the Claude API

Calling Fable 5 from your own code requires nothing more than a standard HTTPS request to the Messages API endpoint. Here is a minimal example using curl, which translates directly to any language or HTTP client:

curl https://api.anthropic.com/v1/messages \   -H “x-api-key: $ANTHROPIC_API_KEY” \   -H “anthropic-version: 2023-06-01” \   -H “content-type: application/json” \   -d ‘{     “model”: “claude-fable-5”,     “max_tokens”: 1024,     “messages”: [       {“role”: “user”,        “content”: “Explain adaptive thinking in two sentences.”}     ]   }

The only change from calling any other Claude model is the model ID. In a production application, you would externalize the API key, add retry handling, and use one of Anthropic’s official SDKs — which brings us to the genuinely new part.

7. What Changes for API Integrations

According to the official Claude API documentation, integrations calling Fable 5 should plan for three changes that do not apply to Opus, Sonnet, or Haiku.

• Refusal handling. Fable 5 includes safety classifiers that can decline certain requests. When that happens on the Messages API, you receive a successful HTTP 200 response with stop_reason: “refusal” — not an error — and the response reports which classifier declined the request. Your error handling should treat this as a distinct case from a rate limit or a server error. You are not billed for a request that is refused before any output is generated.
• Fallback strategy. A refused request can usually be served by another Claude model such as Opus 4.8. The docs list three retry paths: a server-side fallback parameter (in beta), client-side middleware in the official SDKs, or a manual retry you build yourself. Fallback credit refunds the prompt cache cost of switching models, so you do not pay it twice.
• Adaptive thinking. Adaptive thinking is the only mode of thinking in Fable 5. It applies whenever the thinking parameter is unset, and thinking: {“type”: “disabled”} is not supported — so do not port a disabled-thinking configuration from an older integration. Thinking depth is controlled with the effort parameter instead, and the raw chain of thought is never returned.

One more consideration for enterprise teams: Fable 5 offers 30-day data retention and is not available under zero-data-retention agreements. If your compliance posture requires ZDR, you will need to stay on the Opus, Sonnet, or Haiku models for now.

8. Should You Switch to Fable 5?

Not for everything — and that is by design. At double the price of Opus 4.8, Fable 5 is built for the hardest layer of work, not the default layer. A practical routing rule looks like this:

• Use Fable 5 for large migrations, repository-wide refactors, multi-step debugging, deep analytical work, and long agentic sessions where the 1M-token context earns its keep.
• Use Opus 4.8 as the default for heavy day-to-day work — it remains excellent at half the cost.
• Use Sonnet 4.6 or Haiku 4.5 for routine tasks: boilerplate generation, summarization, simple Q&A, and high-volume pipelines.

🎯 INTERVIEW INSIGHT If you are asked in an interview how you would choose between AI models in a production system, the strong answer is escalation-based routing: cheap, fast models handle routine requests, and expensive frontier models are invoked only when task complexity justifies the cost. Claude’s tiered lineup — Haiku, Sonnet, Opus, and now the Mythos-class Fable 5 — is a textbook example of this pattern in practice.

9. Conclusion

Claude Fable 5 is the clearest generational jump Anthropic has shipped in a public model: a new Mythos-class tier, an 11-point lead over Opus 4.8 on SWE-Bench Pro, a more-than-doubled score on production-grade agentic coding, a 1M-token context window, and meaningfully new API behavior around safety refusals and adaptive thinking.

The right move is measured adoption. Try it in the Claude app first, then point it at a genuinely hard task — a migration, a large refactor, or a complex analysis you have been postponing. Keep Opus 4.8 as your daily default, and let Fable 5 earn its price on the work where reasoning quality changes the outcome. The numbers above are launch-week figures, so re-verify pricing, availability, and benchmark replication as the ecosystem catches up.

📘 VERIFY WITH THE OFFICIAL DOCS Claude Fable 5 is only days old, and Anthropic may refine pricing, availability, safeguards, and API behavior over time. For anything you plan to build on, treat Anthropic’s official documentation as the source of truth rather than this article or any third-party summary. The two pages to bookmark are the launch announcement at anthropic.com/news/claude-fable-5-mythos-5 and the API model guide at platform.claude.com. If any detail here ever conflicts with those pages, the official documentation wins — review it directly whenever you have a doubt.

Further Reading

The first two links are Anthropic’s official documentation and should be treated as the authoritative reference for anything in this article. Review them directly whenever you have a doubt.

Anthropic — Claude Fable 5 and Claude Mythos 5 announcement

Claude API Docs — Introducing Claude Fable 5 and Claude Mythos 5

Claude API Docs — Pricing

The Decoder — Anthropic releases Claude Fable 5 and Mythos 5

VentureBeat — Anthropic brings Mythos to the masses with Claude Fable 5