Where is the Fable 5 name from?

Anthropic's model naming uses a two-tier convention. 'Mythos' refers to the unrestricted full-power tier (Mythos Preview, Mythos 5). 'Fable' refers to the public, safeguarded counterpart of the same model — a 'fable' as a softer, story-shaped version of a 'myth.' Both names appeared in Anthropic's June 9, 2026 press release. Future generations may follow the same pattern (Fable 6 / Mythos 6).

How do I access Claude Fable 5?

Claude Fable 5 is available immediately on the Anthropic API (model ID claude-fable-5), claude.ai (Pro/Team/Enterprise plans), Amazon Bedrock, Google Vertex AI, and Claude Code as the default selected model for Pro/Max subscribers. No application is required — same access path as Claude Opus 4.8 and Claude Haiku 4.5. Note that the June 22, 2026 usage-credits update changes how many requests Fable 5 consumes from your claude.ai monthly quota.

Is Claude Fable 5 better than GPT-5.5?

On agentic coding and tool use, yes — by a wide margin. Fable 5 scores 80.3% on SWE-Bench Pro versus GPT-5.5's 58.6%, an 8–10 point lead on Terminal-Bench 2.1 and MCP Atlas. On chat, multimodal, and ChatGPT-ecosystem tasks, GPT-5.5 remains competitive and benefits from ChatGPT's 5M+ Codex user footprint. For autonomous code agents like Claude Code, Cursor, and OpenClaw, Fable 5 is the new state of the art as of June 2026.

Quick Answer

What is Claude Fable 5? Anthropic's Public Mythos Model

Published: June 10, 2026

What is Claude Fable 5?

Claude Fable 5 is Anthropic’s first publicly available Mythos-class model, released June 9, 2026. It’s the most powerful Claude model any developer can access without a vetting process — at roughly double the price of Claude Opus 4.8.

Last verified: June 10, 2026

The 60-second answer

Property	Value
Release date	June 9, 2026
Model ID	`claude-fable-5`
Context window	1M tokens
Max output per request	128k tokens
Input pricing	$10 / million tokens ($5 batch)
Output pricing	$50 / million tokens ($25 batch)
Available on	Anthropic API, claude.ai (Pro/Team/Enterprise), Bedrock, Vertex AI, Claude Code
Companion model	Claude Mythos 5 (`claude-mythos-5`, trusted-access only)
Headline benchmark	SWE-Bench Pro 80.3%, Terminal-Bench 2.1 84.1%

What “Mythos-class” means

Anthropic uses “Mythos” to designate models a tier above the current public flagship (Opus 4.x). Mythos models are trained with significantly more compute and demonstrate frontier capabilities in:

Long-horizon agentic coding
Tool use and MCP orchestration
Scientific reasoning (PhD-level)
Multi-step planning and self-correction

The first Mythos Preview shipped in early 2026 to a limited set of partners and was withheld from the public over cybersecurity concerns (Anthropic warned it could autonomously discover novel vulnerabilities). Fable 5 is the productized public version: same model, with a safeguard layer.

Why two names — Fable and Mythos?

Anthropic explicitly forked the model into two variants on June 9:

Claude Mythos 5 — full, unrestricted model. Available only to vetted partners (~150 enterprise customers, government agencies, cybersecurity firms in a trusted-access program).
Claude Fable 5 — public counterpart. Same base model, plus an automated guardrail layer that refuses responses in high-risk areas (offensive cybersecurity, biology, certain dual-use topics).

Both share pricing, context window, and benchmark scores in non-restricted domains. The only difference is what Fable 5 will refuse to answer.

What can Claude Fable 5 do?

Best-in-class agentic coding

On SWE-Bench Pro (a real-world coding benchmark), Fable 5 scores 80.3% — roughly 9 points above Claude Opus 4.8 and 21 points above GPT-5.5. Tool-use benchmarks (Terminal-Bench 2.1, MCP Atlas) show similar gaps.

This translates to coding agents like Claude Code, Cursor 4, Aider, and OpenClaw completing long-horizon tasks (multi-file refactors, full feature implementations, debugging across services) with substantially fewer human interventions.

Frontier reasoning

GPQA Diamond: 87.8% (PhD-level science questions)
AIME 2025: 96.2% (advanced math competition)
CharXiv Reasoning: 86.6% (chart understanding)

Long context

1M-token context window means you can fit entire mid-sized codebases, full books, or hour-long meeting transcripts in a single request — without chunking.

Same safety profile as Claude

Standard Claude refusals apply on top of the new safeguard layer.

What Claude Fable 5 will refuse

Anthropic’s safeguard layer blocks responses in:

Offensive cybersecurity — exploit development, payload generation, reverse shells, ransomware
Autonomous vulnerability discovery — chained CVE analysis at scale
Biology / bioweapons — synthesis routes, gain-of-function reasoning
Some dual-use research — case-by-case, with documented false positives

Legitimate red-team work, CTF challenges, and academic security research have been hitting false-positive refusals in early testing. For those use cases you need Mythos 5 trusted access.

Coverage: Claude Fable 5 cybersecurity restrictions explained

How much does Claude Fable 5 cost?

Usage	Input	Output
Standard API	$10 / MTok	$50 / MTok
Batch API (50% off)	$5 / MTok	$25 / MTok
Prompt caching (write)	$12.50 / MTok	—
Prompt caching (read)	$1.00 / MTok	—

For comparison:

Opus 4.8: $5 / $25 per MTok (half the price)
GPT-5.5: $5 / $25 per MTok
Gemini 3 Pro: $5 / $25 per MTok
Mythos Preview (deprecated): ~$25 / $125 per MTok

Fable 5 is roughly half the price of the previous Mythos Preview but double the price of Opus 4.8. Use it for the workloads that actually benefit from the marginal performance.

How to access Claude Fable 5

Anthropic API

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": 4096,
    "messages": [{"role": "user", "content": "Hello, Fable 5"}]
  }'

Claude Code

claude --model claude-fable-5
# or set it as default in ~/.claude/config.json

claude.ai

Available in the model picker on Pro, Team, and Enterprise tiers. Free-tier users get Haiku 4.5 only.

Bedrock and Vertex AI

Available immediately via anthropic.claude-fable-5 (Bedrock) and claude-fable-5 (Vertex AI).

The June 22 usage-credits catch

On June 22, 2026, Anthropic activates a new usage-credits system for claude.ai Pro/Max/Team plans. Fable 5 will cost significantly more credits per request than Opus 4.8. If you’re a Pro user expecting to use Fable 5 heavily, the practical monthly quota will be much lower than what you’ve been used to with Opus 4.8.

Coverage: Anthropic Claude Code June 15 billing change

When should you actually use Fable 5?

Scenario	Use Fable 5?
Long agentic coding tasks (multi-file refactor)	Yes
Hard SWE-Bench-style debugging	Yes
PhD-level scientific reasoning	Yes
Routine code edits / chat	No — use Opus 4.8
High-volume RAG / classification	No — use Haiku 4.5
Offensive security research	No — needs Mythos 5 trusted access
Long-context document analysis	Yes (or Gemini 3 Pro)

Sources

Anthropic Newsroom: Claude Fable 5 and Claude Mythos 5 (June 9, 2026)
TechCrunch: Anthropic’s Claude Fable 5 is a version of Mythos the public can access today (June 9, 2026)
llm-stats.com: Claude Fable 5 Review, Benchmarks and Pricing
Vellum AI: Claude Fable 5 & Mythos 5 Full Benchmark Breakdown
TrueFoundry: Claude Fable 5 API, Benchmarks, Pricing
agentpedia.codes: Claude Fable 5 Benchmarks and Prompting Guide
CNBC: Anthropic releases Mythos-like AI model to the public (June 9, 2026)
The Guardian: Anthropic releases ‘safe’ version of Claude Mythos AI model