KVShareArena scores methods that repair KV caches for non-exact-prefix reuse (RAG-retrieved chunks, multi-agent reports) on the fraction of the no-cache-to-full-recompute gap recovered, while computing compute/memory/per-request latency. Key findings: position correction alone suffices unless the query requires multiple sources; paid methods (re-encoding / training) recover 1/2-2/3 of the gap; uncorrected reuse can be worse than no cache; and trained adapters lose more quality than training-free methods when reusing across model checkpoints. Shipped as a pip package with automated submission and a public leaderboard.
KVShareArena benchmarks KV-cache reuse beyond exact prefixes
2026-09-11Frontier watchTRIAL
KVShareArena scores methods that repair KV caches for non-exact-prefix reuse (RAG-retrieved chunks, multi-agent reports) on the fraction of the no-cache-to-full-recompute gap recovered, while computing compute/memory/per-request latency. Key findings: position correction alone suffices unless the query requires multiple sources; paid methods (re-encoding / training) recover 1/2-2/3 of the gap; uncorrected reuse can be worse than no cache; and trained adapters lose more quality than training-free methods when reusing across model checkpoints. Shipped as a pip package with automated submission and a public leaderboard.
Availability: pip package; automated submission; public leaderboard
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
This method family (ReCache, CacheBridge, position repair) has been growing all month without a fair comparison. KVShareArena gives it a leaderboard and two decision rules: don't reuse uncorrected caches across queries needing multiple sources, and prefer training-free repair when crossing checkpoints.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Arxiv | 2609.10266 (Thu 10 Sep digest, announced 2026-09-11T00:00Z) |
|---|---|
| Metric | fraction of no-cache-to-full-recompute gap recovered; compute / memory / per-request latency tracked |
| Findings | position correction sufficient unless multi-source queries; paid methods recover 1/2-2/3 of gap; uncorrected reuse can be worse than no cache; trained adapters degrade more than training-free across checkpoints |
| Availability | pip package; automated submission; public leaderboard |
Technical impact 3
Engineering value 4
Adoption signal 2
Maturity 3
Verification cost 2
Risk 2