Benchmark methodology

Evidence backed benchmarks.

A de-risked approach, backed by multiple reproducible benchmarks — each run head-to-head against standard RAG on the same model, the same questions, and the same retrieval.

Claim by claim

The metric breakdown.

Two measurement families back our data: an accuracy-and-efficiency study at 1,000-document scale on an 8B model, and a fleet-throughput ramp on a 30B model. We keep their attributions separate.

ClaimHow it was measured
91× less text per questionIn a 1,000-document test on an 8B model, Smart CAG re-read only the ~15-token question — where a standard RAG setup re-read ~1,352 tokens of document text, both given the same retrieved documents.
Same answer qualityAcross 500 questions, an independent — and more capable — model graded the answers for meaning and scored the two even: a statistical tie with RAG (details below).
10× faster first tokenIn the 30B fleet test with one request at a time, the first token came back in 28 ms versus 276 ms for RAG.
3.1× throughput / GPU at fleet scaleIn the same 30B test with 24 requests running at once, each GPU finished 9.2 answers per second versus 3.0 for RAG.
$0.14 vs $0.43 per 1k queriesAt that same 24-at-once point: take the box's hourly price and divide by the answers it sustained per hour (formula below).

Fleet-scale throughput

The multi-tenant regime, not a lab best case.

We ramp in-flight concurrency up to 24 on one serving GPU tier and measure both arms on the same model and retrieval. The RAG arm runs in the condition a busy fleet actually creates: a per-request nonce defeats the engine's prefix cache — exactly what other tenants' traffic does under churn — so RAG re-reads its context every request while the resident-KV arm does not.

Setup

Qwen3-30B-A3B (FP8), tensor-parallel across a 4-GPU tier (4× L4, $4.602/hr). Same model and same retrieved documents feed both arms.

Measured, server-side

Time-to-first-token is the real time to the first streamed token. Throughput = completed requests ÷ wall-clock at each concurrency level — queueing included.

Cost formula

GPU $/1k queries = instance $/hr ÷ (sustained queries/sec × 3600) × 1000. The real running price of the box, not a list rate.

ConcurrencySmart CAG q/sRAG q/sCAG TTFTRAG TTFT
43.922.0447 ms296 ms
85.252.4161 ms388 ms
167.552.7380 ms399 ms
249.222.9891 ms648 ms

RAG throughput barely moves as concurrency climbs — it is compute-bound re-reading context on every request — so the resident-KV lead widens with load. Full ramp and the raw run outputs are in the source-of-truth results log.

Answer Quality

Same accuracy, different efficiency

Query answer accuracy is a statistical tie with a standard RAG architecture. On a 1,000-document haystack, resident-KV serving matches an equally-retrieved RAG within noise, while re-processing far less text per question.

Same retriever, both sides

A 1,169-document QASPER haystack, 500 questions, on an 8B model. One shared hybrid retriever (keyword + dense + reranker) feeds both arms — so the comparison isolates the serving path, not retrieval quality.

The tie

A paired significance test put the gap between the two well inside statistical noise — on answer quality, they're not distinguishable.

Independent judge

The result is corroborated by a second, independent frontier judge — a different, more capable model than the one under test, so the verdict doesn't depend on a lenient self-grade.

Reproduce it on your stack.

We'll share the connector, the conformance suite, and the prefix-cache benchmark so your own engineers can run the head-to-head on your traffic shape.

Get in touch