Claude Fable 5.1 Sets Accuracy Record at 67.2%, but Hallucinates More
As seen on the 24/7 Wall St. homepage on September 1, 2026.
Anthropic's newest model sets the accuracy record on this benchmark at 67.2%, but it gets there by answering more questions it should have skipped, which is exactly the reliability trade-off enterprise buyers are pricing.
Claude Fable 5.1 (max) attempts 93.4% of AA-Omniscience questions against Claude Opus 5's 87.8%, and records the highest accuracy we have measured at 67.2%, ahead of Claude Fable 5 at 65.4%. It hallucinates more with this higher attempt rate: of questions it didn’t get correct, https://t.co/tBdCN4DlRM
- Replies1
- Reposts0
- Likes8
Continue ReadingShow less
Claude Fable 5.1 (max) posted the highest accuracy score Artificial Analysis has ever recorded on its AA-Omniscience benchmark, reaching 67.2% against Claude Fable 5's previous mark of 65.4%.
The model gets there partly by attempting more questions, taking on 93.4% of the benchmark's queries. A higher attempt rate means the model is less willing to abstain when uncertain, which is where the reliability trade-off begins to show.
Artificial Analysis notes the model hallucinates more as a direct consequence of that elevated attempt rate. Among questions it did not answer correctly, the failure mode shifts toward confident wrong answers rather than cautious refusals.
For enterprise buyers evaluating which model to deploy in high-stakes settings, that distinction matters more than the headline accuracy number. A model that scores higher by guessing more aggressively demands tighter guardrails.