Search behavior gap: Opus searched in 10/60 answers, GPT models searched every time

santiviquez · x · 2026-11-17

Supplementary data from the API recommendation experiment: with an optional web search tool, usage diverged sharply — Opus 5.5 searched in only 10 of 60 answers (0.8 queries each), Fable 5.1 in 24, while both GPT models searched all 60 times (2.3 and 3.2 queries each).

The author also previously found that enabling web search dropped big-cloud API recommendations from 79% to 60%, while smaller parsing startups rose from 5% to 13%.

Related event: Web Search Changes LLM API Recommendations, Usage Varies by Model(2 posts)→

Original post →

More from Models

Models channel →