Research2026-08-27
Artificial Analysis published a new index that ranks web search services used by AI agents, testing 18 products from 8 providers on three public question-answering tests with the same answer model held constant. Perplexity's medium tier led on combined quality, while Parallel's fast tier was cheapest and quickest per task. All figures come from list prices and a single test harness, so results depend on that setup rather than any particular application.
What changed
No common public comparison of web-search tools for AI agents existed across quality, cost and speed.
What it unlocks
Comparing web search services for AI agents on answer quality, cost per task and time per task using one fixed test setup.
- 18 products across 8 providers
- Perplexity Search (medium) scores 80
- 33 model-only baseline
- $8.41 per 1k tasks cheapest search
Sources