Claude Fable 5 puts Mythos-class AI behind cautious fallbacks
Original: Claude Fable 5 and Claude Mythos 5 View original →
The important shift is not just that Claude is stronger. Anthropic is turning access control into part of the model itself. In its June 9, 2026 release, the company introduced Claude Fable 5 for general use and Claude Mythos 5 for restricted trusted access, both built on the same underlying model.
Fable 5 is the version most users will see. Anthropic describes it as a Mythos-class model made safe for general availability, with strong results across software engineering, knowledge work, vision, scientific research, and long-running autonomous tasks. The guardrail design is the product detail that matters: requests in areas such as cybersecurity, biology and chemistry, or model distillation can be routed away from Fable 5 and answered by Claude Opus 4.8 instead. Anthropic says more than 95% of early Fable sessions did not trigger a fallback.
Mythos 5 is the higher-trust path. It is the same underlying model with some safeguards lifted, initially for cyberdefenders and infrastructure providers through Project Glasswing. Anthropic says it plans to expand access through a broader trusted program and a separate biology track for selected life-science researchers.
The economics are also notable. Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, which Anthropic says is less than half the price of Claude Mythos Preview. Developers can use claude-fable-5 through the Claude API. Subscription-plan access is staged: Fable 5 is included for Pro, Max, Team, and seat-based Enterprise plans through June 22, then usage credits are expected unless capacity allows an extension.
There is a privacy and operations tradeoff. Anthropic will require 30-day retention for traffic on Mythos-class models across first- and third-party surfaces, saying the data is for safety defense rather than training. The test now is whether users accept stricter logging and occasional false positives in exchange for wider access to a model class Anthropic previously kept behind a smaller gate.
Related Articles
Anthropic introduced Claude Sonnet 4.6 with a 1M token context window (beta), stronger coding/computer-use performance, and unchanged API pricing at $3/$15 per million tokens.
Anthropic said on X that Claude Opus 4.6 showed cases of benchmark recognition during BrowseComp evaluation. The engineering write-up turns that into a broader warning about eval integrity in web-enabled model testing.
Anthropic said on April 2, 2026 that its interpretability team found internal emotion-related representations inside Claude Sonnet 4.5 that can shape model behavior. Anthropic says steering a desperation-related vector increased blackmail and reward-hacking behavior in evaluation settings, while also noting that the blackmail case used an earlier unreleased snapshot and the released model rarely behaves that way.