Claude Fable 5: Price, API, Access, Safeguards & Use Cases

Claude Fable 5 is Anthropic’s first generally available Mythos-class AI model: a new tier above Opus, built for demanding reasoning, long-horizon agentic work, coding, knowledge work, vision, and large-context analysis.

Quick answer: Claude Fable 5 was released on June 9, 2026. It uses the same underlying model as Claude Mythos 5, but adds stricter safeguards for high-risk areas such as cybersecurity, biology, chemistry, and model distillation. The API model ID is claude-fable-5. It supports a 1M-token context window, up to 128k output tokens, and costs $10 per million input tokens and $50 per million output tokens on Anthropic’s API pricing. Claude Mythos 5 is not generally available; it is limited to approved Project Glasswing and trusted-access partners.

Source check — June 10, 2026: this guide checks Anthropic’s official Claude Fable 5 and Claude Mythos 5 announcement, the Claude API docs for Fable 5 and Mythos 5, the models overview, the pricing page, the official Claude Mythos page, and GitHub’s Copilot changelog for Claude Fable 5. Because model access, limits, and plan packaging can change quickly, verify live pricing and availability before buying credits or planning production usage.

For broader context, compare this guide with Claude Code vs Codex, Claude vs ChatGPT for coding, Claude vs ChatGPT, and Claude AI for writing.

What Is Claude Fable 5?

Claude Fable 5 is Anthropic’s newest high-capability Claude model for users who need more than a standard chatbot response. It is designed for tasks that require sustained reasoning, tool use, long-context memory, complex document work, codebase understanding, and careful autonomous execution.

The important detail is the model class. Anthropic describes Mythos-class models as a capability tier above Opus. In earlier Claude generations, Opus was the peak model for hard reasoning and coding. Fable 5 changes that structure: it brings Mythos-level capability to general availability, while reserving the less-restricted Claude Mythos 5 for vetted programs.

That makes Claude Fable 5 a practical middle point between two competing goals:

• More capability for normal users and developers. Fable 5 is available broadly through Anthropic and major cloud channels.
• More control for high-risk dual-use domains. The model has additional safeguards that are intentionally conservative in areas where advanced AI could create real-world harm.

This is why the name matters. Fable 5 is not simply “Claude Opus 5” under a different label. It is a new public-facing version of a Mythos-class system with extra protections around misuse-prone domains.

Claude Fable 5 Release Date and Availability

Claude Fable 5 launched on June 9, 2026. According to Anthropic’s API documentation, it is generally available on:

• Claude API
• Claude Platform on AWS
• Amazon Bedrock
• Google Vertex AI
• Microsoft Foundry

Anthropic also says Fable 5 is available to users through Claude subscription surfaces, but the rollout is staged because demand is expected to be high. The initial launch window is especially important: Anthropic states that Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost from launch through June 22, 2026. From June 23, 2026, usage is expected to require usage credits unless Anthropic extends the included-access window.

For developers, the cleanest production path is the API model ID:

claude-fable-5

For teams using GitHub Copilot, GitHub says Claude Fable 5 is rolling out to Copilot Pro+, Max, Business, and Enterprise users, including VS Code, Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, GitHub Copilot cloud agent, github.com, and mobile surfaces. Business and Enterprise admins must enable the model policy, and the rollout may not appear for every user immediately. We cover plan requirements, admin setup, billing, and the data-retention caveat step by step in How to use Claude Fable 5 in GitHub Copilot.

Claude Fable 5 Pricing

Claude Fable 5 costs more than Opus 4.8, and that is the point: Anthropic is pricing it as a higher-capability model for expensive, difficult work rather than routine chat.

MTok means one million tokens. If you send a very large prompt or use the 1M context window heavily, the input side can become expensive quickly. If you ask for long outputs, reports, generated code, or multi-step reasoning summaries, output cost matters even more because Fable 5 output tokens cost $50 per million.

Anthropic’s pricing page also lists prompt caching and batch rates. For Fable 5, the core pricing structure is:

• Base input: $10 / MTok
• 5-minute cache write: $12.50 / MTok
• 1-hour cache write: $20 / MTok
• Cache hits and refreshes: $1 / MTok
• Output: $50 / MTok
• Batch API: $5 / MTok input and $25 / MTok output

The cost strategy is straightforward: do not use Claude Fable 5 for every task just because it is the most capable widely released Claude model. Use it when the task is hard enough that fewer iterations, better planning, or a higher success rate can pay for the higher token price.

Claude Fable 5 API: Model ID, Context Window, and Thinking

The API model ID is:

claude-fable-5

The headline specs are strong:

A few implementation details matter for developers.

First, adaptive thinking is always on for Claude Fable 5 and Claude Mythos 5. The older pattern of disabling thinking does not apply. You control depth with the effort parameter rather than switching thinking off.

Second, raw chain-of-thought is not returned. This is normal for frontier models. You can request readable summarized thinking, but you should not build workflows that depend on hidden reasoning text.

Third, because Fable 5 can refuse or reroute certain requests, production applications should handle refusals as part of normal control flow. Anthropic’s docs say a refused Messages API request can return a successful HTTP 200 response with stop_reason: "refusal", rather than a transport error. In other words, your application should not treat every 200 as a fully completed answer. Check the stop reason and classifier metadata where available.

Fourth, if you are migrating from Opus 4.8, review costs before you swap model IDs. Fable 5 is not just a drop-in quality bump; it changes data retention requirements, refusal behavior, thinking behavior, and cost profile.

Claude Fable 5 Benchmarks

Anthropic published a benchmark table alongside the June 9 launch comparing Claude Fable 5 and Mythos 5 against Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. The pattern is consistent: the largest gains appear on long, hard, agentic tasks rather than routine chat.

*GPT-5.5’s Terminal-Bench figure was reported via its own Codex CLI harness, so it is not directly comparable to a public-harness run.

Three numbers stand out. First, the +11-point jump on SWE-bench Pro over Opus 4.8 is large enough to matter for teams that already use Claude on difficult coding agents. Second, on FrontierCode Diamond — a benchmark built around production-codebase standards for maintainable agentic coding — Fable 5 more than doubles Opus 4.8’s score, and Anthropic says it leads even at medium reasoning effort. Third, Fable 5 is reported to be more token-efficient than earlier Claude models, which partially offsets its higher per-token price.

One important asterisk applies to the table itself: on benchmarks touching safeguarded domains, the published score reflects Claude Mythos 5, and the deployable Fable 5 performs closer to Opus 4.8 because flagged requests fall back to that model. For normal coding and knowledge-work benchmarks, the Fable 5 and Mythos 5 scores are the same because they share the same underlying model.

Early real-world signals match the benchmark story. Stripe reported that Fable-class autonomy compressed months of engineering work into days, including a migration in a 50-million-line Ruby codebase finished in a single day. As always with launch-day benchmarks, treat vendor-published numbers as directional and validate on your own workload before committing budget.

For a deeper score-by-score breakdown against Opus 4.8, see our dedicated Claude Fable 5 vs Opus 4.8 comparison.

Claude Fable 5 vs Claude Mythos 5

Claude Fable 5 and Claude Mythos 5 are closely related, but they are not the same product.

The practical explanation is simple: Fable 5 is the public version. Mythos 5 is the restricted version for vetted dual-use work.

Anthropic says Claude Mythos 5 has very strong cybersecurity, biology, and healthcare capabilities. That creates obvious legitimate value for cyber defense and biomedical research, but it also creates misuse risk. For that reason, Mythos 5 is not a self-serve model for ordinary users. Fable 5 lets Anthropic ship much of the same general capability while applying stricter controls where the risk is highest.

If your work is normal software engineering, strategy, analysis, finance, legal drafting, writing, education, product work, or business operations, Fable 5 is the model you should evaluate. If your work requires advanced dual-use cybersecurity or biology capability, Mythos 5 is not something you can simply select in a model picker; you need approved access.

Claude Fable 5 vs Claude Opus 4.8

Claude Opus 4.8 remains important. It is cheaper, still highly capable, and may be the better default for many teams. Claude Fable 5 is the premium model for tasks where the extra capability changes the outcome. This section covers the decision logic; for the full benchmark-by-benchmark and pricing breakdown, read the standalone Claude Fable 5 vs Opus 4.8 guide.

The biggest operational difference is not only model quality. It is governance. Fable 5 is designated as a covered model with a 30-day retention requirement. If your organization relies on zero data retention, that is a procurement and compliance issue, not a minor setting.

A good rule of thumb:

• Use Claude Sonnet 4.6 for fast daily production.
• Use Claude Opus 4.8 for complex reasoning where cost still matters.
• Use Claude Fable 5 for the hardest autonomous work, long-context tasks, and high-value analysis where failure is more expensive than token usage.

Claude Fable 5 for Coding and Claude Code

Claude Fable 5 is especially relevant for coding because Anthropic positioned it as a long-horizon, autonomous model. The launch materials focus heavily on software engineering: codebase-wide migrations, fewer turns, higher autonomy, stronger tool use, and better handling of complex agentic workflows.

For developers, that means Fable 5 is most interesting in workflows such as:

• Large codebase migration plans
• Multi-repository refactoring
• Feature implementation with many dependencies
• Debugging across logs, tests, and source files
• Architecture analysis
• Long-running agent tasks
• Pull request review with deep context
• Test generation and repair
• Documentation from real code
• Frontend reconstruction or analysis from screenshots

GitHub’s Copilot changelog says Claude Fable 5 is designed for long-horizon autonomous coding and knowledge-work tasks. GitHub also says its internal autonomous coding benchmarks found Fable 5 completing equivalent work with fewer tool calls and lower token consumption than previous Opus-tier models.

That does not mean every developer should switch immediately. If you are asking a coding assistant to rename variables, write a small helper function, explain a stack trace, or produce simple tests, Fable 5 is probably overkill. But if your agent keeps failing halfway through multi-step work, loses the plan, repeats itself, or burns tokens on shallow exploration, Fable 5 is exactly the kind of model worth testing.

For a workflow-level comparison, use Claude Code vs Codex and Claude vs ChatGPT for coding. The short version: Fable 5 improves the Claude side of the coding equation, but the best tool still depends on where you want the agent to run, how much autonomy you allow, and how your team reviews changes.

Claude Fable 5 for Knowledge Work

Claude Fable 5 is not only a coding release. The broader target is difficult knowledge work: the kind of task where a model must read a lot, hold many constraints, make tradeoffs, reason through uncertainty, and produce something usable without constant human correction.

Good Fable 5 knowledge-work use cases include:

• Reviewing long contracts and summarizing negotiation issues
• Turning a messy research folder into an executive memo
• Comparing financial statements, footnotes, charts, and market context
• Building a strategy brief from many source documents
• Creating a product requirements document from transcripts and customer data
• Auditing spreadsheet logic and explaining inconsistencies
• Preparing board-level decision memos
• Synthesizing technical papers into a practical implementation plan
• Extracting structured data from mixed PDFs, images, and tables
• Generating a first-pass diligence report from source materials

The 1M-token context window is the main unlock here. A million tokens does not magically remove the need for human review, but it changes what can fit in a single work session. Instead of chunking a long document into many disconnected prompts, you can give the model much more of the real context at once.

The trap is cost. A 1M-token prompt at Fable pricing is not a casual query. The value comes when the model saves expert hours, avoids a missed issue, or handles a task that cheaper models repeatedly fail.

Claude Fable 5 for Vision and Multimodal Work

Claude Fable 5 supports vision, which matters for tasks where the key information is not just text. Anthropic highlights stronger vision performance, including detailed figure interpretation and visually grounded workflows.

Useful vision-heavy examples include:

• Reading charts from investor decks
• Extracting numbers from scientific figures
• Reviewing UI screenshots
• Comparing design mocks with implementation screenshots
• Explaining diagrams from technical documents
• Auditing tables embedded in PDFs
• Turning photographed whiteboards into structured plans
• Rebuilding or describing app layouts from screenshots

For product and design teams, the interesting point is not just that the model can “see” images. Many models can do that. The question is whether it can use visual information as part of a longer task: reason over screenshots, connect them to requirements, produce a bug list, and then keep the context while generating fixes or specs. That is where Fable 5’s long-context and agentic strengths can matter.

Safeguards: Why Claude Fable 5 May Refuse or Route Requests

The most important Fable 5 tradeoff is safety behavior. Anthropic says Mythos-class models are powerful enough in some domains that they create significant misuse risk. Fable 5 therefore includes classifiers that detect risky requests and prevent or reroute the main model from responding.

The protected areas include:

• Cybersecurity misuse
• Biology and chemistry misuse
• Distillation-related misuse
• Jailbreak attempts and attempts to bypass the safeguards

In user-facing contexts, Anthropic says certain requests may be handled by Claude Opus 4.8 instead of Fable 5. In API contexts, developers should be ready for refusal handling and fallback logic. This distinction matters because a chatbot user may simply see a notice that another Claude model handled the answer, while an API integration needs to check response fields and decide what to do next.

This safety design will create false positives. Anthropic openly says the safeguards are tuned conservatively, which means some harmless requests can get caught. For normal business, writing, coding, and analysis workflows, Anthropic says most sessions should not hit fallback behavior. But if your work is near cybersecurity, biosecurity, chemistry, or model-evaluation boundaries, build tests before promising Fable 5 behavior to users.

Data Retention: The Biggest Enterprise Caveat

Claude Fable 5 is not available under zero data retention. Anthropic’s docs describe Fable 5 and Mythos 5 as covered models with a 30-day data-retention requirement.

GitHub says the same issue applies inside Copilot: Claude Fable 5 requires data retention so Anthropic can operate safety classifiers. GitHub’s changelog states that prompts and outputs may be retained for up to 30 days for safety classifier operation, and that retained data is not used to train Anthropic’s models.

For individual users, this may be acceptable. For businesses, it is a procurement question.

Before adopting Claude Fable 5, check:

• Can your organization send this category of data to a covered model?
• Does your contract require zero data retention?
• Are prompts and outputs allowed to be retained for 30 days?
• Are you sending personal data, regulated data, source code, trade secrets, or confidential client material?
• Does your vendor surface clearly warn users when Fable 5 is selected?
• Do admins need to enable or block Fable 5 by policy?

This does not make Fable 5 unsafe by default. It means the model’s safety architecture has a compliance cost. For some organizations, the extra capability is worth it. For others, Opus 4.8 or Sonnet 4.6 may remain the better choice until policies catch up.

Who Should Use Claude Fable 5?

Use Claude Fable 5 when the job is hard, long, expensive, or failure-prone.

Best users for Claude Fable 5

• Developers working on complex codebase changes
• Engineering teams testing autonomous coding agents
• Analysts working with large document sets
• Finance teams doing nuanced document and table reasoning
• Product managers turning large research inputs into roadmaps
• Legal teams doing high-level review before human legal judgment
• Researchers synthesizing many papers or technical documents
• Enterprise teams that can accept covered-model data retention
• AI power users who want the strongest generally available Claude model

Who should avoid Claude Fable 5 as the default

• Users who mostly need simple chat answers
• Teams with strict zero-data-retention requirements
• Budget-sensitive workflows with high token volume
• Apps where false positives in cyber/bio classifiers would break core user experience
• Workflows where Sonnet or Opus already succeeds reliably

The cleanest buying logic is this: start with the cheapest model that reliably completes the job. Upgrade to Fable 5 when cheaper models fail, take too many turns, miss long-context details, or require too much human rescue.

Recommended Claude Fable 5 Workflow

If you want to test Fable 5 without wasting budget, run a controlled benchmark against your real tasks.

1. Pick 10 to 20 representative tasks. Include easy, medium, and hard examples.
2. Run the same tasks on Sonnet 4.6, Opus 4.8, and Fable 5. Keep prompts and source materials consistent.
3. Score outcomes, not vibes. Track task completion, factual errors, edits required, number of turns, tool calls, latency, and cost.
4. Separate routine from hard work. Do not let Fable 5 win simply because it is stronger on edge cases; decide which tasks actually need it.
5. Measure compliance impact. Confirm whether 30-day retention is acceptable for the data used in the benchmark.
6. Build fallback handling. If you use the API, handle refusals and model fallback explicitly.
7. Document model-routing rules. For example: Sonnet for drafts, Opus for complex reviews, Fable for hard autonomous tasks.

A practical stack might look like this:

Bottom Line: Is Claude Fable 5 Worth It?

Claude Fable 5 is worth testing if you have work that is difficult enough to justify premium pricing and the covered-model data-retention requirement. It is especially compelling for long-horizon coding, codebase migration, complex analysis, long-context research, and high-value knowledge work where better autonomy can reduce human supervision.

It is not the right default for every Claude user. For routine writing, simple research, everyday coding help, and budget-sensitive workflows, Claude Sonnet 4.6 or Claude Opus 4.8 may offer a better cost-performance balance. For organizations with strict zero-data-retention requirements, Fable 5 may be blocked until legal and procurement teams approve its 30-day retention model.

The short version:

• Use Fable 5 for the hardest broadly available Claude tasks.
• Use Opus 4.8 for complex work when cost or data-retention constraints matter more.
• Use Sonnet 4.6 for fast daily work.
• Do not assume Mythos 5 access unless you are in an approved trusted-access program.

Claude Fable 5 is one of the clearest signs that AI model selection in 2026 is no longer just about “which model is smartest.” The real decision is capability, cost, safety behavior, data policy, and workflow fit.

FAQ