Anthropic Fable 5
Anthropic introduced Fable 5 as its first 'Mythos-class' model made safe for general use — a new tier whose capabilities exceed anything the company had previously released to the public, sitting above the Opus class. It claimed state-of-the-art results on nearly all tested benchmarks, with standout performance in software engineering, knowledge work, vision, and scientific research, citing examples such as a customer compressing a large-codebase migration from months into a day. The broad release was made possible by new safeguards that block responses in sensitive areas (primarily cybersecurity and biology), with blocked queries routed to a Claude Opus 4.8 response instead.
Fable 5 launched June 9, 2026 on the Claude API and consumption-based Enterprise plans, and was included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans through June 22, after which subscription access requires usage credits; API pricing is $10 per million input tokens and $50 per million output tokens. Mythos 5 is the same underlying model with safeguards lifted in some areas, restricted to a small group of vetted cyberdefenders and infrastructure providers and initially deployed through Project Glasswing in collaboration with the US government.
It's early — Fable 5 is barely two weeks old, so this is a first read, not a settled reality. The headline 'best model we've shipped' claim has held up better than most launch hype: independent trackers corroborate the top-tier story, with Artificial Analysis placing Fable 5 at #1 on its Intelligence Index (64.9, ~5 points ahead of GPT-5.5) and third-party harnesses showing ~95% on SWE-Bench Verified, while practitioner reaction to hard, multi-file and long-running agentic coding has been unusually strong. Caveats are already visible: Anthropic's flashiest number (80.3% on SWE-Bench Pro) came from its own scaffolding rather than a neutral harness and is not yet independently reproduced, the model silently falls back to Opus 4.8 on a meaningful share of queries, and at $10/$50 per million tokens it costs roughly twice Opus 4.8. The more consequential story for executives is governance, not raw capability: within 48 hours Anthropic was forced to walk back an undisclosed capability limit buried in its system card ('we made the wrong tradeoff, and we apologize'), Microsoft reportedly limited employee use over data-retention terms, and a US export-control directive restricted Fable/Mythos access by nationality. What remains unseen: durable real-world reliability at scale, neutral third-party results on the hardest benchmarks, and whether the policy-gated Mythos tier holds as a precedent or proves leaky.
Fable 5 makes the most capable tier explicitly conditional on clearance rather than budget — Mythos-class power is gated to vetted cyberdefenders and, via export controls, withheld by nationality — so for leaders the live question shifts from 'can we afford the frontier?' to 'are we permitted to use it, and on what data terms?' The early data-retention and silent-safeguard episodes mean capability and governance now have to be evaluated together, not separately.