On June 9, 2026, Anthropic launched Claude Fable 5, its first publicly available model in the top-tier "Mythos-class," across all platforms. It is available the same day in Claude Code and on the Claude Platform.
June 9, 2026 · Anthropic
Claude Fable 5: Anthropic ships its most powerful public model — days after warning AI is "too dangerous"
The first broadly released "Mythos-class" model lands worldwide on Claude Code and the Claude Platform — claiming SOTA in software engineering, knowledge work, science and vision, with guardrails baked into the public version.
91/100
Senior-Engineer benchmark score for Fable 5
<5%
Sessions auto-routed back to Opus 4.8 on high-risk queries
0.5–1M
Tokens consumed per task — heavy and slow
Senior-Engineer benchmark — scored to scale
Score out of 100 · taller column = higher score
Two configurations, one model
Claude Fable 5
General release · safety version
Worldwide on Claude Code & Platform, same day
High-risk queries auto fall back to Opus 4.8 (<5%)
Claude Mythos 5
Unrestricted version · same model weights
Project Glasswing trusted partners only
Aimed at defensive cyber & biomedical research
High-risk safety routing
Queries in sensitive domains never reach the new model.
Cyber · bio/chem · distillation query
→
Auto fallback to Opus 4.8
→
User notified
Strengths
Called the strongest coding model yet
Fixes bugs across an entire large codebase
One-shot 3D game or ~2-min animation
Suits long-running agents & complex reports
Concerns
Slow and heavy — overkill for everyday work
Pricing ~$10 in / $50 out per M tokens — ~2× Opus
Opus 4.8 seen as enough for smaller use cases
Some users unhappy with public guardrails
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…
