88Z AI, 10 June — Anthropic has released its most powerful model, Claude Fable 5, which outperforms previous generations across the board and leads significantly in benchmarks like SWE-Bench Pro. The model offers higher token efficiency, capable of migrating a 50-million-line codebase in one day. Additionally, Anthropic launched the unguarded version, Claude Mythos 5, available exclusively to Glasswing partners. For safety, Claude Fable 5 includes guardrails that downgrade sensitive requests to Claude Opus 4.8.
At a glance
- Claude Fable 5 outperforms predecessors with high token efficiency, migrating 50 million lines of code in a day.
- Unguarded Claude Mythos 5 is exclusive to Glasswing partners, with sensitive requests handled by Opus 4.8.
The update
Anthropic has officially launched Claude Fable 5, its most powerful AI model to date. The new model surpasses its predecessors in all performance metrics, showing significant leadership in professional benchmarks such as SWE-Bench Pro. Beyond strong reasoning capabilities, Claude Fable 5 greatly improves token efficiency, capable of migrating a codebase of up to 50 million lines within 24 hours, demonstrating exceptional engineering processing power. At the same time, Anthropic released Claude Mythos 5, a version without safety guardrails, which is currently available exclusively to Glasswing partners. To ensure a safe experience for mainstream users, the standard Claude Fable 5 retains strict safety guardrails. When the system detects sensitive or high-risk requests, it automatically downgrades and transfers the processing task to the Claude Opus 4.8 model, thereby ensuring security while delivering powerful performance.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 机器之心Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.