Back to developer mode
Research bake-off
The same question through Gemini grounded search and Perplexity structured output, side by side.
These are not the same task
Gemini is asked for at most four sentences of prose and returns character offsets, so it can mark the exact clause a source supports. Perplexity is asked to fill a schema and returns a flat source list with no offsets. Judge on: is the headline answer correct, latency, cost, source quality, and whether the structure helps or gets in the way.
ShapeTurning the planner off and pinning the shape it chose isolates the planner hop's latency cost.
Gemini
Grounded search, prose with inline citations
No run yet.
Perplexity
Structured output against a fixed schema
No run yet.