Keiro /search/lite vs Parallel turbo: the cheap-endpoint benchmark · Keirolabs<br>NEW/search/lite is live at $0.25 / 1k reqs, beating Parallel turbo 91/100See the battle →
Start FreeTry in Postman
91/100<br>Outright wins
67%<br>Composite share (341.6 vs 170.4)
6/9<br>Archetypes swept clean (Parallel scored zero)
$0.25/1k<br>Per 1,000 requests
The contenders<br>Same 100 queries. Same intent. Same 15 results. Different output.<br>Both endpoints get the same query and the same intent string, with result counts equalized at 15. The difference is what comes back: Keiro returns scoped results, real code, authority sources, and an image SERP. Parallel returns turbo excerpts with no scope knobs.<br>Keiro · lite<br>endpointPOST kierolabs.space/api/v2/search/lite<br>authAuthorization: Bearer $KEIRO_KEY<br>requestcount 15 · prompt · content:true<br>result cap15 results<br>scope knobssite · market · time · type=images<br>The page body text per result, scope controls, real code blocks, and an image SERP. The knobs restrict results to a domain, region, date window, or image type.<br>$0.25 / 1k requests
Parallel · turbo<br>endpointPOST api.parallel.ai/v1/search<br>authx-api-key: $PARALLEL_KEY<br>requestmax_results 15 · objective · mode: turbo<br>result cap15 results<br>scope knobsnone exposed<br>Turbo excerpts per result. No site, market, time, or image knobs. Web-text only, no image SERP.<br>see Parallel pricing
Composite points<br>341.6 vs 170.4. A 2× composite lead.<br>Each query contributes weighted points across the quality metrics that apply to it (see How it's scored). Keiro takes 67% of the composite and 91 of 100 outright wins. Parallel's 9 wins cluster in scope-domain, where its broader SERP edges Keiro's hard site= filter, plus a few general-knowledge queries.<br>100200300Keiro /search/lite341.6Parallel turbo170.4
91Keiro outright wins<br>9Parallel outright wins<br>0Ties
By archetype<br>Keiro sweeps 8 of 9 archetypes.<br>Recall, code, authority, multi-modal, and both locale and date scope all go to Keiro clean. Parallel takes only scope-domain, where its broader SERP edges Keiro's hard site= filter on tight domains. Share is Keiro's slice of the weighted composite in each archetype; the depth figure is a secondary signal, not the score.<br>passagen=16<br>15/1wins K/P<br>66%<br>A specific fact lives in the page body, not the snippet.<br>EU AI Act fine tiers<br>depth 4.9×
yieldn=16<br>14/2wins K/P<br>84%<br>Deep content retrieval for RAG ingestion.<br>how RAG works<br>depth 4.6×
recalln=14<br>14/0wins K/P<br>70%<br>Multi-entity, multi-facet coverage.<br>React vs Vue vs Svelte<br>depth 5.4×
scope-domainn=10<br>4/6wins K/P<br>36%<br>Parallel's only win. Keiro's site= knob narrows harder but returns less body on tight domains.<br>RAG on arxiv<br>depth 0.6×
scope-localen=10<br>10/0wins K/P<br>64%<br>market= enforces a regional SERP. Parallel has no region knob.<br>AI reg Germany<br>depth 6.0×
scope-daten=4<br>4/0wins K/P<br>88%<br>time=w|m restricts to a window. Parallel has no date knob.<br>AI announcements past week<br>depth 3.4×
coden=12<br>12/0wins K/P<br>71%<br>Real code blocks returned, not prose mentions of code.<br>useEffect AbortController<br>depth 5.4×
authorityn=12<br>12/0wins K/P<br>71%<br>Official, gov, edu, and primary-source domains surfaced.<br>GDPR lawful basis<br>depth 5.4×
imagesn=6<br>6/0wins K/P<br>100%<br>type=images returns image results. Parallel is web-text only.<br>Vision Pro photos<br>image SERP, not chars
Where the quality comes from<br>What Keiro returns that Parallel can't.<br>Some are capabilities Parallel does not expose: scope knobs, an image SERP, real code blocks, authority sources. Some are the quality outcomes those capabilities produce: the answer is present in the payload, and every entity in a multi-facet query is covered. Each maps to a knob or response shape, with the example query that exercises it.<br>Passage presencepassage 15/1<br>The expected answer actually appears in the returned payload, not just on a related page the SERP surfaced. Scored on whether the answer terms turn up.<br>e.g. GDPR max fine amount
Recallrecall 14/0<br>Multi-entity and multi-facet queries come back with every entity covered, not just the top hit. Scored on entities and facets hit.<br>e.g. React vs Vue vs Svelte
Scope conformancelocale 10/0 · date 4/0<br>Results stay inside the requested domain, region, or date window. market= and time= are enforced; Parallel exposes no scope knobs.<br>e.g. AI reg Germany → market=de
Real code blockscode 12/0<br>Code queries return runnable code with real markers, not a prose sentence that mentions the function name.<br>e.g. useEffect AbortController
Authority sourcesauthority 12/0<br>Official, government, edu, and primary-source domains are surfaced for regulatory and standards queries.<br>e.g. GDPR lawful basis
Image SERPimages 6/0<br>type=images returns image results for visual queries. Parallel is web-text only and returns no image SERP at all.<br>e.g. Vision Pro photos
Depth is the mechanism, not the score<br>Depth is what puts the answer in the payload.<br>Keiro body chars (y) vs Parallel body chars (x), both log-scaled. Above the dashed parity line = Keiro returned more body text, which is why its...