Fable 5 API Guide — Model ID, Pricing & Code Examples

This guide covers everything you need to call Fable 5 through the Anthropic API: the exact model ID, pricing, the parameters that changed, and working code.

The essentials

• Model ID: claude-fable-5 (use this exact string — no date suffix)
• Endpoint: POST https://api.anthropic.com/v1/messages
• Pricing: $10 / million input tokens, $50 / million output tokens
• Context: 1M tokens · Max output: 128K tokens (streaming required for large outputs)

Quick start (TypeScript)

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic(); // reads ANTHROPIC_API_KEY

const response = await client.messages.create({
  model: "claude-fable-5",
  max_tokens: 16000,
  thinking: { type: "adaptive" },
  messages: [{ role: "user", content: "Explain quantum entanglement simply." }],
});

for (const block of response.content) {
  if (block.type === "text") console.log(block.text);
}

Quick start (Python)

from anthropic import Anthropic

client = Anthropic()  # reads ANTHROPIC_API_KEY

response = client.messages.create(
    model="claude-fable-5",
    max_tokens=16000,
    thinking={"type": "adaptive"},
    messages=[{"role": "user", "content": "Explain quantum entanglement simply."}],
)
print(response.content[0].text)

Parameters that changed on Fable 5

Fable 5 keeps the same API surface as Claude Opus 4.7/4.8, with one extra restriction:

| Parameter | Status on Fable 5 | | --- | --- | | thinking: {type: "adaptive"} | ✅ Recommended | | thinking: {type: "enabled", budget_tokens: N} | ❌ Returns 400 | | thinking: {type: "disabled"} | ❌ Returns 400 — omit the param instead | | temperature / top_p / top_k | ❌ Removed, returns 400 | | Last-assistant-turn prefill | ❌ Returns 400 — use structured outputs | | output_config: {effort: "low" \| "medium" \| "high" \| "xhigh" \| "max"} | ✅ Supported |

Effort guidance: use xhigh for coding and agentic work, high for most intelligence-sensitive tasks, and low/medium for fast, scoped tasks.

Controlling cost

At $10/$50 per MTok, cost discipline matters:

1. Prompt caching — cache reads cost ~10% of base input price. Put stable content (system prompt, documents) before the cache breakpoint.
2. Batch processing — the Batches API runs at 50% of standard prices for non-urgent workloads.
3. Task budgets (beta) — give agentic loops a token budget the model is aware of and self-moderates against.
4. Route by difficulty — send routine traffic to Opus 4.8 or Sonnet 4.6; reserve Fable 5 for tasks that need the ceiling. See our comparison.

Try before you integrate

Don't want to set up an API key just to evaluate it? Chat with Fable 5 in our playground — free credits on sign-up.

---

fable-5.net is an independent resource and is not affiliated with Anthropic. Always check the official Anthropic documentation for the latest API details.