Unverified50% confidenceBenchmarkExact time
In simulated inference scenarios with low memory and long context, KV sharing brings three improvements: up to 31.4% higher cache hit rate, up to 23.6% higher throughput, and up to 83.6% lower p90 time to first token.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/23/2026
First Seen
Valid until: 11/6/2026
Sources
Related Entities
Related Claims
UnverifiedThe model offers three reasoning levels: low, high and max; in the author's tests max generated about 16% more tokens than high, with almost the same total time for reasoning tasks.70% similarUnverifiedExperiments by the Long-Term team showed that using the same model with only adjustments to rule constraints and the memory system raised the score from 52.8% to 66.5%, jumping the ranking from Top 30 to Top 5.69% similarUnverifiedIn the composite score for memory retrieval, similarity has the largest weight, recency accounts for a small proportion, and the result is multiplied by an importance factor with a range of approximately 0.8 to 1.269% similarUnverifiedAfter introducing an explicit resolver to handle information supersession relationships, stale-memory failures dropped to zero and overall accuracy jumped from 33% to 78%.68% similarUnverifiedIn the ablation study, when no data is used and only parametric memory is relied on, even the best model performs poorly; after providing early-quarter data, performance improves by about 25 percentage points; but there is still a 12-16 percentage point gap from using the latest data.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088250API
curl https://kongchang.com/api/v1/knowledge/claims/1088250MCP
get_claim(id=1088250)