Artificial Analysis launched a Search Index, and running with no search at all scores 33.

18 search API products across 8 providers, scored as the equal-weighted mean of DeepSearchQA F1, BrowseComp exact-answer accuracy and AA-Omniscience accuracy. Perplexity medium leads at 80 for $91.39 per 1,000 tasks at 28.6s each; Parallel advanced and Brave LLM-context tie at 75; the model-only baseline scores 33 at $2.92 per 1,000 tasks. That spread is the first public number putting a price on retrieval quality, and 33 is higher than I'd have guessed for a model answering from memory alone. (Artificial Analysis)