Model scorecard
Perplexity sonar-deep-research
Composite 56.8/100 across quality, throughput, latency, price efficiency and usable context — measured through the gateway, not vendor-reported.
B56.8/100perplexity/sonar-deep-researchtext · 128k context
Capability
eval compositeQuality (blended eval)87
Category: reasoning
Usable context6
128k advertised window
Serving
gateway measuredThroughput17
60 output tokens/s median
Time to first token0
640ms p50
Economics
at provider costPrice efficiency56
$8.00 per 1M output tokens
Quality per dollar32
llmcloud adds $0 to these token prices.
Suggested usage
Not a default pick today. It is a reasonable explicit choice when you need this vendor specifically, or as a failover target inside auto:reasoning.
→ Full workload matrix