Claude Fable 5

https://www.anthropic.com/claude/fable

$10 per million input tokens / $50 per million output tokens (API); included for Pro/Max/Team/Enterprise (platform credits)

SWE-bench Pro 80.3% — #1 coding performance as of July 2026
1M Token Context Window — full monorepo comprehension, no re-explaining
128K Max Output Tokens — sustained generation for long tasks
Multi-Day Autonomous Sessions — state persists between sessions
Self-Verification Loop — writes and runs its own tests, iterates on failures
Sub-Agent Delegation — coordinates parallel agent workflows
Vision Capability — inspects UIs, diagrams, screenshots against requirements
Mythos-Class Architecture — Anthropic's most advanced model tier, made safe for general use
Available on Claude Code, Claude API, AWS Bedrock, Google Vertex AI, Microsoft Foundry, GitHub Copilot, GitLab Duo
Safety Classifiers — declines/reshapes unsafe requests; audit-ready for enterprise governance
$10/$50 per million tokens — premium pricing justified for complex engineering work

80.3% SWE-bench Pro — best-in-class coding performance as of July 2026
1M token context window — full monorepo comprehension in a single session
Multi-day autonomous sessions with state persistence between sessions
Self-verification loop — writes and runs its own tests without prompting
Sub-agent delegation — coordinates parallel agent workflows
Vision capability — inspects UIs, diagrams, and screenshots against specs
Export-control review passed — global access restored July 1, 2026
Broad platform availability — Claude Code, API, AWS Bedrock, Vertex AI, Foundry, GitHub Copilot, GitLab Duo

Premium pricing: 2x Opus 4.8 — overkill for routine edits and simple tasks
Safety classifiers occasionally decline requests competitors accept
Requires agent harness (Claude Code, custom build) for full value — not a chat-first experience
Knowledge cutoff January 2026 — not current to today