Two random search APIs match the best of 12 in agent search benchmark
Dry-Golf-5690 · reddit · 2026-10-29
A Reddit post shares agent search API benchmark numbers: 12 providers, the same 363 queries, metric = known answer in top-5 results.
Key data: the single best provider hit 68%; two randomly picked providers combined hit 77%; three random 80%; all 12 84%. The surprising finding is that two random providers (68%) essentially match the best single one — so having any fallback matters more than obsessively benchmarking your primary.
Notably, 16% of queries had no answer from any provider, which vendor-swapping can't fix — that's on how the agent searches or whether it reads pages. The author asks whether others prefer fallback-on-miss or fanning out to multiple providers every time.
More from coding & agent
- Anthropic probes Claude Code usage complaints, finds no regression, ships targeted token fixes — patience_cave · 2026-10-29
- Google Gemini rolls out a central hub for custom skills and connected apps — sysrpl · 2026-10-29
- New Prompt Optimization Beats GEPA by ~14% and Estimates Performance Bounds — sanmikoyejo · 2026-10-29
- New Chinese book 'Codex: From Beginner to Master' targets hands-on Codex learners — lxfater · 2026-10-29
- SJTU's TAO Cuts LLM Agent Serving Latency Up to 26.8% With Tool-Aware Joint Scheduling — AIOR-Research · 2026-10-29
- Monopoly game fully written by Claude in a single day — sysrpl · 2026-10-29