Researchers at UC Santa Barbara and Hanbat National University published CoinRAG on August 7, 2026, a KV cache optimization for retrieval-augmented generation that replaces chunk-level reuse with sub-chunk "information nugget" caching. On LongBench multi-hop question answering benchmarks, it achieves 5.3% relative F1 improvement — 41.7 vs. 39.6 — against Standard RAG and Cache-Augmented Generation baselines, while holding prefill inside a P99 time-to-first-token budget of 100 ms.

SystemLongBench F1Relative F1 GainP99 TTFT Budget
Standard RAG (baseline)39.6Not reported
Cache-Augmented Generation (baseline)39.6Not reported
CoinRAG41.7+5.3%≤ 100 ms
FIG. 02 CoinRAG vs. baseline systems on LongBench multi-hop QA (F1 score, P99 TTFT ≤ 100 ms) — arXiv 2608.07458 (UC Santa Barbara / Hanbat National University, 2026)

Cache-Craft (arXiv 2502.15734) quantifies the problem in production: prefill consumes 55.4% of total inference time on one cluster (Sys-X) and 76% on another (Sys-Y). Redundant chunk processing costs over 12 billion tokens per month on LLaMA-3-70B across eight A100 GPUs — roughly $50,000 in GPU time. Retrieved chunks contain answers alongside substantial irrelevant context, so caching the full chunk caches the noise along with the signal.

MetricValueContext
Prefill share of total inference time — Sys-X55.4%One production cluster
Prefill share of total inference time — Sys-Y76.0%Second production cluster
Redundant chunk tokens per month>12 billionLLaMA-3-70B, 8 × A100 GPUs
Estimated GPU cost of redundant prefill~$50,000 / monthLLaMA-3-70B, 8 × A100 GPUs
FIG. 03 Production-scale prefill costs quantified by Cache-Craft (arXiv 2502.15734) — Cache-Craft, arXiv 2502.15734

CoinRAG caches below the chunk level. Offline, an LLM extracts text-span-based "information nuggets" from each chunk — compact units isolating relevant assertions. KV caches for these nuggets are precomputed and stored. At inference, two-stage retrieval narrows candidates at chunk level, then selects query-relevant units within those chunks. The system assembles a contextualized KV representation by composing precomputed nugget caches with a chunk-level context signal, giving the model cross-chunk coherence without encoding full chunks.

CoinRAG two-stage retrieval: nuggets are precomputed offline; at inference, chunk-then-nugget retrieval assembles a composed KV representation without re-encoding full chunks.
FIG. 04 CoinRAG two-stage retrieval: nuggets are precomputed offline; at inference, chunk-then-nugget retrieval assembles a composed KV representation without re-encoding full chunks. — arXiv 2608.07458

Chunk-level systems face a trade-off: hit the 100 ms time-to-first-token SLA with fewer chunks or exceed it to answer multi-hop questions. CoinRAG shifts that frontier by making each retrieval unit smaller and more semantically dense. The same latency budget fits more relevant evidence because each nugget uses fewer tokens — and fewer prefill FLOPs — than the chunk it came from.

Competing approaches exist. CacheBlend (EuroSys '25) reuses any precomputed chunk cache regardless of position and selectively recomputes KV values for a small subset of tokens to restore cross-chunk attention, achieving 2.2–3.3× time-to-first-token reduction over full recompute. It ships as LMCache, an open-source vLLM integration. CacheClip uses an auxiliary model to identify which tokens to recompute, reaching 3.33× prefill speedup while retaining 85.2–91.1% of full-attention performance. CoinRAG avoids recomputation by precomputing at finer grain from the start.

SystemCaching GranularityTTFT / Prefill ImprovementQuality RetentionRequires Fine-tuningOpen-source Integration
CacheBlend (LMCache)Chunk-level; selective KV recompute for cross-chunk attention2.2–3.3× TTFT reduction vs. full recomputeRestored via selective recomputationNoYes — vLLM (LMCache)
CacheClipToken-level; auxiliary model selects tokens to recompute3.33× prefill speedup85.2–91.1% of full-attention performanceNo (aux model only)Not specified
CoinRAGSub-chunk nuggets; no recomputation at inferenceWithin 100 ms P99 TTFT on multi-hop QA+5.3% F1 over baseline (41.7 vs. 39.6)Yes — nugget-aware fine-tuning requiredNot specified
FIG. 05 Comparison of KV cache reuse approaches for RAG prefill acceleration — arXiv 2608.07458; EuroSys '25 (CacheBlend); arXiv 2510.10129 (CacheClip)

Trade-offs exist. Nugget extraction requires an offline LLM pass over every document before serving. Nugget-aware fine-tuning means the target model is no longer off-the-shelf. Deployment teams on high-churn corpora face cache invalidation pressure — every document update triggers re-extraction and re-caching. The paper evaluates on three LongBench multi-hop QA datasets; single-hop factoid retrieval and summarization tasks are not covered.

CoinRAG suits workloads where multi-hop QA dominates, document churn is low, and one fine-tuning run is feasible. CacheBlend (via LMCache) offers lower barrier entry for existing vLLM stacks with mixed retrieval patterns. The 5.3% F1 gain under a hard 100 ms P99 budget is measurable, but the fine-tuning dependency makes it a system commitment.