Artificial Analysis ranks OpenAI Web Search at 74 on Search Index

OpenAI Web Search debuted on the Artificial Analysis Search Index with a score of 74, marking a 41-point increase over GPT-5.6 Luna without search (33). The result places OpenAI as the 5th provider on the index and 7th out of 26 tested products, tying with You.com (highlights), Nimble (standard), and Exa (auto), behind Perplexity variants (77 to 80), Octen (77), Parallel advanced, and Brave LLM context (75).

On individual evaluation suites, OpenAI Web Search recorded its highest ranking on the factual benchmark AA-Omniscience, scoring 72% accuracy to rank 3rd of 26 variants, one point behind Firecrawl at 73%. On BrowseComp, which tests multi-hop reasoning across sequential searches, it reached 73.5% and placed 13th, trailing Octen and Perplexity variants that scored between 85% and 87%.

Tested within a single Responses API call using GPT-5.6 Luna with medium reasoning and medium search context, OpenAI Web Search billed roughly 40k input tokens per task including search results, compared to 125k for Perplexity low. Overall task cost averaged approximately $0.05, lower than 17 of 25 Search API configurations on the leaderboard. OpenAI charges $10 per 1,000 web_search calls, with retrieved search content billed as model input tokens.