Anthropic’s Claude Opus 5 offers lower cost, fewer restrictions than Fable 5

Anthropic split Claude Fable 5 and Claude Mythos 5 on June 9, 2026, turning one new large language model into two products with different access rules. Fable 5 was the general-use release, while Mythos 5 kept the same underlying model but lifted safety blocks for a small set of trusted partners through Project Glasswing.

Anthropic’s system card said Fable 5 added safeguards that blocked high-risk biology and cybersecurity tasks. The company described Fable 5 as its most capable widely released model, built for demanding reasoning and long-horizon agentic work, and said it could run for days. Mythos 5 removed those safeguards, but Anthropic kept it in limited release rather than opening it broadly.

That split matters because price now sits beside safety as a buying criterion. One pricing summary put Fable 5 at $10 per million input tokens and $50 per million output tokens, and said that sticker price was twice Claude Opus 4.8, which Anthropic introduced on May 28, 2026. For enterprises comparing vendors, that gap pushes procurement teams toward a simple calculation: pay more for the newest model with tighter controls, or stay on a cheaper prior release when the workload does not justify the premium.

The tradeoff is sharper for developers building agentic systems and for government users working through compliance reviews. Fable 5’s guardrails on biology and cybersecurity lower the risk of misuse, but they also narrow what the model can do on request. Mythos 5 went the other direction, offering a more permissive version of the same model, yet only to trusted partners, which means the least restricted option was also the hardest to get.

Anthropic later said access to Fable 5 and Mythos 5 was restored on July 1 after a June 12 US government-related issue referenced in its redeploying notice. The company had already released the model publicly after a prior private rollout, underscoring how quickly access, pricing and safety policy now shape which frontier models actually reach day-to-day deployment.