# Claude Fable 5

Claude Fable 5 is Anthropic's most capable widely released model, pairing a context window of 1M tokens with always-on adaptive thinking, configurable effort, and safeguards built for broad availability.

- **Model ID:** `anthropic/claude-fable-5`
- **Type:** chat
- **Providers:** anthropic, bedrock, vertexAnthropic
- **Context window:** 1,000,000
- **Maximum output tokens:** 128,000
- **Pricing:** $10/1M input tokens, $50/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/claude-fable-5

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'anthropic/claude-fable-5',
  prompt: 'Why is the sky blue?'
})
```

## About

Claude Fable 5 became available on AI Gateway on July 1, 2026. Access was briefly suspended under US export controls and restored in July 2026 once the controls were lifted; the restored model is the same model, with updated and more robust safety classifiers. Anthropic describes Claude Fable 5 as a Mythos-class model made safe for general availability, with capabilities that exceed any model Anthropic had widely released before. The launch materials report leading results across nearly every benchmark Anthropic tested, spanning software engineering, knowledge work, vision, and scientific research.

Long-horizon agentic work is the design target. Claude Fable 5 handles long-running, complex, and asynchronous tasks where previous models needed more frequent check-ins. At launch, Claude Fable 5 supports the memory tool, code execution, programmatic tool calling, compaction, tool result clearing through context editing, and vision input, plus task budgets in beta. Those are the building blocks for agents that run for hours without losing state.

Adaptive thinking is always on. Claude Fable 5 decides when and how deeply to reason, and `thinking: { type: 'disabled' }` is not supported. Use the `effort` parameter to control thinking depth and cost per request. Raw chain of thought is never returned: set `thinking.display` to `summarized` for a readable summary, or keep the default `omitted`.

Safety classifiers ship as part of Claude Fable 5 and can decline certain requests, including occasional false positives on routine tasks such as coding and debugging. A refusal returns `stop_reason: "refusal"` as a successful response, not an error, and reports which classifier declined. Configure fallback models through the `models` array in `providerOptions.gateway` so AI Gateway retries refused requests on the next Claude model in your list, such as Claude Opus 4.8 or Claude Sonnet 5. Fallbacks work on every API format.

Through AI Gateway, Claude Fable 5 is available with the standard unified API, observability, retries, and failover across Anthropic, Bedrock, Google Vertex AI. Set the model to `anthropic/claude-fable-5` in the AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python. AI Gateway mirrors provider pricing with no markup and adds no platform fee on inference, including for Bring Your Own Key requests.

## What to consider

Claude Fable 5 includes safety classifiers that can decline certain requests, and even routine tasks such as coding and debugging can occasionally trigger them. A declined request returns `stop_reason: "refusal"` as a successful response rather than an error. Configure fallback models with `providerOptions.gateway.models` so AI Gateway retries refused requests on the next model in your list, such as Claude Opus 4.8 or Claude Sonnet 5. Fallbacks work on every API format, and AI Gateway observability tracks refusals.

Adaptive thinking is the only thinking mode on Claude Fable 5. Use the `effort` parameter to control reasoning depth and cost. Premium per-token rates reward routing: send the hardest work to Claude Fable 5 and keep high-volume simple traffic on a cheaper Claude tier.

## When to use

### Best For

- **Long-Horizon Agentic Work:** Long-running, complex, and asynchronous tasks that previously needed more frequent check-ins
- **Demanding Reasoning Problems:** Hard software engineering, knowledge work, and scientific research where output quality justifies premium compute
- **Large-Context Workloads:** A context window of 1M tokens and up to 128K tokens output tokens cover big codebases and long agent histories
- **Stateful Autonomous Agents:** Memory, compaction, code execution, and programmatic tool calling keep multi-hour sessions on track
- **Production Agent Infrastructure:** Workflows that benefit from AI Gateway routing across Anthropic, Bedrock, Google Vertex AI with retries and failover

### Consider Alternatives When

- **Cost-Sensitive Production Traffic:** Claude Sonnet 5 delivers strong agentic performance at a much lower per-token price
- **Latency-Critical Interactive Use:** A Haiku-tier model is faster and cheaper for well-bounded, high-throughput requests
- **Simpler Response Handling:** Sonnet-tier models don't include Fable-specific safety classifiers, which keeps refusal handling out of your integration

## Best for

- **Long-Horizon Agentic Work:** Long-running, complex, and asynchronous tasks that previously needed more frequent check-ins
- **Demanding Reasoning Problems:** Hard software engineering, knowledge work, and scientific research where output quality justifies premium compute
- **Large-Context Workloads:** A context window of 1M tokens and up to 128K tokens output tokens cover big codebases and long agent histories
- **Stateful Autonomous Agents:** Memory, compaction, code execution, and programmatic tool calling keep multi-hour sessions on track
- **Production Agent Infrastructure:** Workflows that benefit from AI Gateway routing across Anthropic, Bedrock, Google Vertex AI with retries and failover

## Consider alternatives

- **Cost-Sensitive Production Traffic:** Claude Sonnet 5 delivers strong agentic performance at a much lower per-token price
- **Latency-Critical Interactive Use:** A Haiku-tier model is faster and cheaper for well-bounded, high-throughput requests
- **Simpler Response Handling:** Sonnet-tier models don't include Fable-specific safety classifiers, which keeps refusal handling out of your integration

## Frequently asked questions

### What is Claude Fable 5 built for?

Claude Fable 5 is Anthropic's most capable widely released model, built for the most demanding reasoning and long-horizon agentic work. Claude Fable 5 handles long-running, complex, and asynchronous tasks where previous models needed more frequent check-ins.

### How does thinking work on Claude Fable 5?

Adaptive thinking is the only thinking mode, and Claude Fable 5 decides how deeply to reason per request. Control depth and cost with the `effort` parameter. Raw chain of thought is never returned; set `thinking.display` to `summarized` for a readable summary.

### What happens when Claude Fable 5 declines a request?

The response returns `stop_reason: "refusal"` as a successful response, not an error, and reports which classifier declined. Configure fallback models with `providerOptions.gateway.models` and AI Gateway retries the request on the next model in your list, such as Claude Opus 4.8 or Claude Sonnet 5, on every API format.

### What agent tooling does Claude Fable 5 support?

At launch, Claude Fable 5 supports the memory tool, code execution, programmatic tool calling, compaction, tool result clearing through context editing, and vision input. Effort control ships as standard, and task budgets are available in beta.

### What context window and output limit does Claude Fable 5 have?

Claude Fable 5 has a context window of 1M tokens and supports up to 128K tokens output tokens per request, enough for large codebases, long document sets, and extended agent histories.

### Does AI Gateway support Zero Data Retention for Claude Fable 5?

Zero Data Retention is not currently available for this model. Anthropic does not offer Zero Data Retention for Claude Fable 5 because some misuse patterns are only visible across cumulative requests. Prompts and completions are retained for 30 days and are not used to train Claude. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

### What is the pricing for Claude Fable 5?

AI Gateway mirrors provider pricing with no markup and adds no platform fee on inference. Rates are listed on this page and shift when providers update their pricing.

### How do I call Claude Fable 5 through AI Gateway?

Set the model to `anthropic/claude-fable-5` in the AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python. AI Gateway handles authentication, retries, and failover across `anthropic`, `bedrock`, `vertex`.

## Links

- [Model page](https://vercel.com/ai-gateway/models/claude-fable-5)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://www.anthropic.com/news/claude-fable-5-mythos-5)