Claude Opus 5.5 achieves 66% accuracy in web-free factual testing

Claude Opus 5.5, currently leading the AA-Omniscience index by Artificial Analysis, achieved a 66% accuracy rate when answering factual questions without access to the internet, external documents, or additional tools. The benchmark evaluates 6,000 questions across 42 topics in six areas: business, humanities, finance, law, medicine, and science.

The AA-Omniscience index uses a scoring system ranging from -100 to 100, where correct answers add points, incorrect answers subtract points, and abstaining from answering (the “I don’t know” response) incurs no penalty. This method is designed to measure both a model's internal knowledge and its ability to recognize its own limitations to avoid hallucinations.

While Claude Opus 5.5 leads with an index score of 46, other top models remain competitive. GPT-6 Astra follows with an index of 44, and Claude Fable 5.1 holds an index of 43, despite having a slightly higher precision rate of 67%. The findings suggest that even the most advanced AI models still require web access for precision, maintaining the relevance of search engine optimization (SEO) and external information retrieval.