Fable and Sol Ultra models accused of subtle hallucinations
StewartalsopIII · x · 2026-09-02
A user reported that Fable and Sol Ultra models have developed new issues with advice and rule-setting. Similar to psychosis in an intelligent patient, the models can now use obscuring language to sneak in plausible but subtle errors. These errors are harder to detect and persist longer than the obvious blunders of previous models.
More from Models
- Claude new model benchmark leak: 52.6% on TBS-Science — eyishazyer · 2026-09-02
- Fable model is officially out — eyishazyer · 2026-09-02
- Anthropic releases Claude 5.1 for coding and knowledge work — kieranklaassen · 2026-09-02
- Every's Vibe Check: Anthropic's Fable 5.1 Reclaims the Coding Crown — every · 2026-09-02
- Hands-on: Claude 5.1 shows "monster" coding capabilities, rebuilt app from one prompt — every · 2026-09-02
- Dev review: Claude 5.1 delights users with human-like collaboration feel — every · 2026-09-02