Gemini Omni coming soonLearn more

Claude Fable 5 API

Access Claude Fable 5 through EvoLink with model ID `claude-fable-5`. Use one API key and explicit routing to send only your hardest coding, agent, and long-context requests to the top Claude tier.
Price: 

$9.000(~ 612 credits) per 1M input tokens; $45.000(~ 3060 credits) per 1M output tokens

$11.250(~ 765 credits) per 1M cache write tokens; $0.900(~ 61.2 credits) per 1M cache read tokens

Web search tool charged separately per request.

Highest stability with guaranteed 99.9% uptime. Recommended for production environments.

Use the same API endpoint for all versions. Only the model parameter differs.

Claude Fable 5 API — Anthropic's Most Intelligent Model

Route Anthropic's most capable widely released model through EvoLink for the hardest coding, long-horizon agent, and long-context work — where stronger reasoning is worth a premium over Opus.

Claude Fable 5 API visualization

Is Claude Fable 5 the right route for your hardest tasks?

Reach for it on the hardest cross-file and architecture work

When a task needs the strongest available planning — reasoning across a large codebase, redesigning architecture, untangling a difficult refactor, or producing a high-stakes migration plan — Fable 5 is the frontier route that sits above Opus 4.8.

Claude Fable 5 coding workflows

Use it for long-horizon agents where failure costs the most

When an agent runs long tool loops, carries heavy state across many steps, and a wrong turn is expensive to recover from, Fable 5 works as the top escalation route — not the default for every agent request.

Claude Fable 5 agent workflows

Use it for the largest-context, highest-stakes analysis

When one reasoning path must absorb large repos, long specs, logs, and research packs and the conclusion really matters, Fable 5 is best suited to context-heavy work where you want the strongest synthesis available.

Claude Fable 5 long-context analysis

When should you route to Claude Fable 5, and when is Opus 4.8 enough?

This product page is not a review. It helps teams decide which requests justify the top tier — Fable 5 costs about twice as much as Opus 4.8 — and which requests should stay on Opus 4.8 or a lighter Claude model.

Keep Opus 4.8 as the strong default

Opus 4.8 already handles most high-value coding, agent, and long-context work well. Treat Fable 5 as a deliberate escalation for the hardest requests, not as a blanket upgrade for all Claude traffic.

Send only frontier-difficulty requests to Fable 5

Reserve Fable 5 for tasks where Opus 4.8 leaves quality on the table — the most complex codebase work, the longest agent runs, and the highest-stakes long-context decisions where stronger reasoning pays for itself.

Fable 5 costs about 2x Opus 4.8 — route deliberately

Because Fable 5 sits at a higher price point, the routing question is whether the extra reasoning is worth the premium for each request. Keep everyday and lighter traffic on Opus, Sonnet, or Haiku where they fit.

Claude Fable 5 and Opus 4.8: what is different?

Use this as a routing decision table, not a full benchmark. Keep Opus 4.8 as the strong default, and escalate only your hardest, highest-value requests to Fable 5 — it costs about twice as much.

Decision pointKeep Opus 4.8 when...Escalate to Fable 5 when...
Existing workloadOpus 4.8 already meets production quality on this route.The task is at the edge of what Opus 4.8 can do reliably.
Claude CodeChanges are scoped and Opus 4.8 lands them cleanly.The task spans the whole repo, hard architecture, or risky migrations.
Agent workflowThe agent loop is stable and recoverable on Opus 4.8.The run is long, heavily stateful, and failure is expensive to undo.
Long contextCurrent long-context prompts work reliably on Opus 4.8.The decision rides on the largest repos, specs, and research packs.
Cost vs qualityYou want the best value for high-volume premium traffic.The extra reasoning is worth roughly 2x the price on this request.

How do you call the Claude Fable 5 API through EvoLink?

Create your EvoLink key, use `claude-fable-5` as the model ID, and route deliberately with caching, the effort parameter, and clear model selection.

1

Step 1 - Create one unified API key

Sign up for EvoLink and use one API key to manage Claude model access instead of maintaining separate integration logic for each provider.

2

Step 2 - Use claude-fable-5

Set the model parameter to `claude-fable-5` so this request enters the Fable 5 route explicitly instead of being mixed into generic Claude traffic.

3

Step 3 - Choose a routing strategy by workflow

Decide when a request truly needs Fable 5 and when to route back to Opus 4.8, Sonnet, or Haiku based on task difficulty, context size, latency, and cost.

How do 1M context, 128K output, and caching work together on Fable 5?

The real production question is not a single parameter. It is how long context, output capacity, reasoning depth, and repeated-request cost shape your routing policy at the top tier.

Context

Use 1M context for high-value long tasks

You can process large documents, research packs, or codebases in one request, but long context does not mean every request should carry the maximum possible context.

Capacity

Use 128K output for complete plans and long code

Longer output helps with code, plans, reports, and structured deliverables, but production prompts should still set clear output boundaries to avoid waste.

Intelligence

Control depth with adaptive thinking and effort

Fable 5 uses adaptive thinking and decides how much to reason per request. Use the `effort` parameter to trade thoroughness against token spend instead of a fixed thinking budget. Verify supported effort levels on the current API documentation.

Multimodal

Use vision input for screenshots and document review

Combine text and image inputs for screenshot analysis, document review, UI inspection, and multimodal debugging workflows.

Efficiency

Prompt caching for repeated context

Use cache writes and cache hits for stable prompts and recurring long inputs instead of sending the same context as a fresh request every time. The official cache hit rate is much lower than full input price, so caching matters more at the top price tier.

Reliability

Keep routing explicit for future migration

Keep `claude-fable-5` explicit so teams can compare Fable, Opus, Sonnet, and Haiku routes across quality, latency, and cost.

Switch Claude routes inside the same API

EvoLink gives you access to Claude models through one API. Use Fable 5 for frontier-difficulty paths, keep Opus 4.8 as the strong default, and route everyday high-frequency requests to Sonnet or Haiku when they fit better. All models share the same EvoLink API endpoint, so you can switch models with one parameter.

Plan your Claude Fable 5 rollout

These guides split the API, how-to, and comparison intent so this product page can stay focused on access, model ID, and pricing.

Claude Fable 5 API access FAQ

Everything you need to know about the product and billing.

Claude Fable 5 supports a 1M token context window and up to 128K output tokens in a single request, which makes it suitable for large codebases, long documents, and analysis-heavy workflows.
Anthropic lists Claude Fable 5 at $10 per million input tokens and $50 per million output tokens, with separate prompt caching rates. EvoLink pricing may reflect your account, credit, or route configuration, so use the pricing table on this page for the current EvoLink rate.
Use `claude-fable-5` in the model field when routing this model through EvoLink.
Opus 4.8 is the strong default for most high-value coding, agent, and long-context work. Route to Fable 5 only for frontier-difficulty requests where stronger reasoning meaningfully improves the result — the hardest codebase work, the longest agent runs, and the highest-stakes long-context decisions.
Only for the right requests. Fable 5 input and output pricing is roughly double Opus 4.8, so it pays off when the extra reasoning prevents costly errors or correction loops. For everyday work, keep traffic on Opus 4.8 or lighter Claude models.
Fable 5 uses adaptive thinking and decides how much to reason per request. At launch, Anthropic documents that `budget_tokens` is not supported and that sampling parameters such as temperature and top_p are also not supported. Use the `effort` parameter to control reasoning depth and token spend. Verify current parameter support, supported effort levels, and thinking behavior on the Anthropic API documentation, as these details may evolve.
The `effort` parameter helps you trade off response thoroughness against token usage and latency. Fable 5 supports higher effort levels than previous Claude models, which is useful when correctness matters more than cost on frontier tasks. Check the Anthropic API documentation for the current list of accepted values.
Anthropic lists Claude Fable 5 as generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry beginning June 9, 2026. On EvoLink, use the model page and API docs to confirm the current route and account availability.
Yes. Claude Fable 5 supports text and image inputs, which makes it useful for screenshots, visual documents, and multimodal analysis workflows.
Refer to Anthropic's official models overview for the authoritative knowledge and training-data cutoff dates, since they can be updated as the model is documented.
Anthropic says Claude Fable 5 has safeguards for higher-risk cybersecurity, biology, chemistry, and distillation requests. Depending on the request and configuration, the response may be blocked or handled by Claude Opus 4.8. Treat this as an important production caveat for sensitive workflows.
No self-serve EvoLink route should be assumed for Claude Mythos 5. Anthropic describes Mythos 5 as limited availability through Project Glasswing and approved customer channels, while Claude Fable 5 is the generally available Mythos-class route.
No. Fable 5 is the top tier and is meant for the hardest requests. Most teams keep Opus 4.8 as the default and route simpler work to Sonnet or Haiku, reserving Fable 5 for the highest-value, highest-difficulty workloads.