Claude Fable 5 for stock trading

Fable 5 had a five-day public window. Anthropic launched it on 2026-06-09 as the most capable Claude model ever shipped to general API users, then pulled it on 2026-06-12 after a US export-control directive landed. If your agent ran on Fable 5 during those days, this page exists so you can document what was actually under the hood.

The headline numbers: 1 million token context, 128K output per request, multimodal, $10 per million input tokens, $50 per million output tokens. Twice the price of Opus 4.8 on both sides. Anthropic positioned it as the long-horizon agentic option, optimized for software engineering and vision-heavy work. The same weights ship as Claude Mythos 5; the Fable label carried tighter guardrails on cybersecurity, biology, and chemistry prompts, with automatic fallback to Opus 4.8 when those gates fired.

For trading, the practical implication is that Fable 5 was overkill for tick-by-tick decisions and well-suited to research and strategy generation. MeanStreak's setup was the obvious pattern: Fable 5 generates and refines the strategy logic offline, then a cheaper Claude model (Sonnet or Haiku) runs the trade loop with that logic compiled in. Operators who used Fable in the trade loop directly burned through their daily token budget fast.

The export-control suspension is open-ended. Anthropic has not committed to a return date, and the published export-licensing process implies months at minimum. If you were running on Fable 5 when it went dark on 2026-06-12, the cheapest like-for-like fallback is Opus 4.8. If you can drop a half-step in capability, Sonnet 4.6 will do most of the same work for a tenth the cost.