benchmarks/web search for coding agents/most token-efficient web search api for coding agents
independently benchmarked · 100 questions · updated Sep 26, 2026

Most Token-Efficient Web Search API for Coding Agents, 2026

Most Token-Efficient Web Search API for Coding Agents: TinyFish leads 5 search APIs at 6.15 points per 1,000 tokens, benchmarked on 100 coding tickets in 2026.

What this page compares. This page ranks measured results from 2 boards: coding-agent documentation tickets (search and page fetch) and coding-agent documentation tickets (search only).

How to read the result. The ranking divides the score by tokens returned, so shorter results that still answer rank higher. 7 configurations below 80% of the best score are left out, so nothing ranks on speed or price alone.

Coding-agent documentation tickets, search and page fetch, ranked by accuracy per 1,000 tokens

Every row ran the same 100 questions. Provider names link to the official docs for the configuration measured.

RankProviderTask completionMean search timeMedian task timeMedian cost / taskMedian task tokensCompletion / 1k tokensList price
3Perplexitysearch_context_size=high77.7%991 ms21.5 s$0.08320,0623.87$5.00 / 1k queries
4Exa deeptype=deep83.0%4.0 s36.8 s$0.1323,6603.51$12.00 / 1k queries
5Exa autotype=auto81.7%1.2 s22.7 s$0.1127,4332.98$7.00 / 1k queries
6Parallel advancedmode=advanced77.0%3.1 s33.4 s$0.09627,0922.84$5.00 / 1k queries
7Parallel basicmode=basic76.0%1.6 s30.0 s$0.1132,8092.32$5.00 / 1k queries

Coding-agent documentation tickets, search only, ranked by accuracy per 1,000 tokens

RankProviderTask completionMean search timeMedian task timeMedian cost / taskMedian task tokensCompletion / 1k tokensList price
2Perplexitysearch_context_size=low77.3%957 ms18.3 s$0.0598,7658.82$5.00 / 1k queries
3Parallel fastmode=fast66.7%953 ms22.1 s$0.05412,4605.35$1.00 / 1k queries
4Parallel turbomode=turbo64.7%333 ms18.7 s$0.05914,1304.58$1.00 / 1k queries
5Exa fasttype=fast66.3%626 ms20.0 s$0.1022,3442.97$7.00 / 1k queries

How the ranking was measured

Questions answered by this comparison

Which provider leads most Token-Efficient Web Search API for Coding Agents?

TinyFish leads this table at 6.15 points per 1,000 tokens, on 100 coding tickets.

How was this measured?

The same coding agent worked the same documentation tickets with one search API at a time, three independent runs per ticket. A ticket passes only when the patch checks out and cites a source the agent actually retrieved in that run. Search-only rows return results; search-and-fetch rows can also read the page. Search time, total task time, task tokens and the median cost per task (model plus API) are reported beside completion.

How should speed and cost be read beside the score?

TinyFish recorded 24.3 s median task time and $0.050 median cost per task. The most accurate configuration is rarely the fastest or the cheapest, so each is reported in its own column.

Do vendors pay to be ranked?

No. Inclusion and rank do not depend on vendor payments. Every API used the same questions, answering model and scoring.

What does this page not test?

This page ranks measured results from 2 boards: coding-agent documentation tickets (search and page fetch) and coding-agent documentation tickets (search only).